Modell-Release-Übersicht
Aufklappbare Übersicht der KI-Modell-Releases der letzten 18 Monate mit tagesgenauen Ergänzungen und Context-Angaben, soweit auffindbar.
Prompt
Mach mir bitte in agentic-workflow-design eine Übersichts-HTML-Seite mit den Model-Releases der letzten 18 Monate. Releases hänge ich an. Bitte ergänze zu jedem Model der letzten 6 Monate den genauen Tag und wenn auffindbar die maximale context-size. Bau die Seite bitte so, dass sie pro Quartal und Monat auf/zuklappbar ist. Gestalte sie gerne etwas ansprechend. Erstelle ein neues Verzeichnis dafür. Nachtrag: Kimi, GLM-5.1, GLM-4.7 und weitere Zwischen-Releases ergänzen; gerne mit Generator statt riesiger Inline-Liste. Danach die Badges unter dem Datum entfernen.
Antwort
Ergebnis: statische HTML-Seite unter model-release-overview/index.html, später auf Generator und JSON-Datenquelle umgestellt. Die Übersicht wuchs von 101 auf 125 Releases, inklusive Kimi/Moonshot-, GLM/Z.ai-, Qwen-, Grok-, OpenAI- und DeepSeek-Ergänzungen. Verifiziert wurden Generatorlauf, Sortierung, Duplikate sowie Desktop- und Mobile-Rendering.
Ergebnis
HTML anzeigen
<!doctype html>
<html lang="de">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Model-Releases 2025-2026</title>
<style>
:root {
color-scheme: light;
--bg: #f6f3ec;
--paper: #fffdf8;
--ink: #17201d;
--muted: #62706a;
--line: #ded7c8;
--line-strong: #c9bfac;
--teal: #0e766e;
--teal-soft: #dcefeb;
--amber: #d4922f;
--amber-soft: #f6e5c8;
--rose: #a84f61;
--blue: #466d91;
--shadow: 0 18px 46px rgba(39, 32, 22, 0.12);
--radius: 8px;
font-family: Inter, ui-sans-serif, system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", sans-serif;
}
* { box-sizing: border-box; }
body {
margin: 0;
background:
linear-gradient(135deg, rgba(14, 118, 110, 0.08), transparent 28rem),
linear-gradient(315deg, rgba(212, 146, 47, 0.12), transparent 34rem),
var(--bg);
color: var(--ink);
letter-spacing: 0;
}
a { color: var(--teal); text-decoration-thickness: 0.08em; text-underline-offset: 0.18em; }
a:hover { color: #084f4a; }
.page {
width: min(1180px, calc(100vw - 32px));
margin: 0 auto;
padding: 42px 0 56px;
}
header {
display: grid;
grid-template-columns: minmax(0, 1fr) 360px;
gap: 28px;
align-items: end;
padding: 22px 0 30px;
border-bottom: 1px solid var(--line-strong);
}
h1 {
margin: 0;
max-width: 820px;
font-size: clamp(2.25rem, 5vw, 4.7rem);
line-height: 0.94;
letter-spacing: 0;
font-weight: 780;
}
.intro {
margin: 18px 0 0;
max-width: 760px;
color: var(--muted);
font-size: 1.04rem;
line-height: 1.55;
}
.stats {
display: grid;
grid-template-columns: repeat(2, minmax(0, 1fr));
gap: 10px;
}
.stat {
background: rgba(255, 253, 248, 0.78);
border: 1px solid var(--line);
border-radius: var(--radius);
padding: 14px 15px;
box-shadow: 0 8px 24px rgba(39, 32, 22, 0.06);
}
.stat b {
display: block;
font-size: 1.72rem;
line-height: 1;
margin-bottom: 6px;
}
.stat span {
color: var(--muted);
font-size: 0.78rem;
text-transform: uppercase;
letter-spacing: 0.08em;
font-weight: 720;
}
.toolbar {
position: sticky;
top: 0;
z-index: 5;
display: grid;
grid-template-columns: minmax(0, 1fr) auto auto;
gap: 10px;
align-items: center;
margin: 22px 0;
padding: 12px;
background: rgba(246, 243, 236, 0.92);
border: 1px solid var(--line);
border-radius: var(--radius);
backdrop-filter: blur(14px);
}
input[type="search"] {
width: 100%;
border: 1px solid var(--line-strong);
border-radius: 6px;
background: var(--paper);
color: var(--ink);
font: inherit;
font-size: 0.96rem;
padding: 10px 12px;
outline: none;
}
input[type="search"]:focus {
border-color: var(--teal);
box-shadow: 0 0 0 3px rgba(14, 118, 110, 0.14);
}
button {
border: 1px solid var(--line-strong);
border-radius: 6px;
background: var(--paper);
color: var(--ink);
padding: 10px 13px;
font: inherit;
font-size: 0.88rem;
font-weight: 720;
cursor: pointer;
}
button:hover { border-color: var(--teal); color: var(--teal); }
.legend {
display: flex;
flex-wrap: wrap;
gap: 8px;
margin: 0 0 20px;
color: var(--muted);
font-size: 0.88rem;
}
.pill {
display: inline-flex;
align-items: center;
min-height: 24px;
border: 1px solid var(--line);
border-radius: 999px;
padding: 3px 9px;
background: rgba(255, 253, 248, 0.72);
color: var(--muted);
font-size: 0.78rem;
font-weight: 720;
white-space: nowrap;
}
.pill.day { background: var(--teal-soft); color: #09534f; border-color: #badbd5; }
.pill.context { background: var(--amber-soft); color: #7b4f12; border-color: #e6c88f; }
.pill.source { background: #ece8df; color: #594f42; }
.quarter {
margin: 16px 0;
background: rgba(255, 253, 248, 0.68);
border: 1px solid var(--line-strong);
border-radius: var(--radius);
box-shadow: var(--shadow);
overflow: clip;
}
summary {
cursor: pointer;
list-style: none;
}
summary::-webkit-details-marker { display: none; }
.quarter > summary {
display: flex;
justify-content: space-between;
gap: 16px;
align-items: center;
padding: 18px 20px;
background: linear-gradient(90deg, rgba(14, 118, 110, 0.12), rgba(255, 253, 248, 0.6));
border-bottom: 1px solid transparent;
}
.quarter[open] > summary { border-bottom-color: var(--line); }
.quarter-title {
display: flex;
align-items: baseline;
gap: 12px;
min-width: 0;
}
.quarter-title strong {
font-size: 1.23rem;
line-height: 1.2;
}
.quarter-title span,
.quarter-meta {
color: var(--muted);
font-size: 0.9rem;
}
.month {
margin: 12px;
border: 1px solid var(--line);
border-radius: var(--radius);
background: rgba(255, 253, 248, 0.86);
overflow: clip;
}
.month > summary {
display: flex;
justify-content: space-between;
gap: 12px;
align-items: center;
padding: 13px 14px;
background: #fffaf0;
border-bottom: 1px solid transparent;
}
.month[open] > summary { border-bottom-color: var(--line); }
.month-name {
font-weight: 790;
font-size: 1rem;
}
.rows {
display: grid;
gap: 0;
padding: 0;
}
.release {
display: grid;
grid-template-columns: 166px minmax(220px, 0.9fr) minmax(160px, 0.6fr) minmax(0, 1.5fr);
gap: 16px;
align-items: start;
padding: 15px 14px;
border-bottom: 1px solid var(--line);
position: relative;
}
.release:last-child { border-bottom: 0; }
.release::before {
content: "";
position: absolute;
left: 98px;
top: 0;
bottom: 0;
width: 1px;
background: var(--line);
}
.date-cell {
display: flex;
flex-direction: column;
gap: 7px;
align-items: flex-start;
position: relative;
}
.dot {
width: 11px;
height: 11px;
border: 2px solid var(--paper);
border-radius: 50%;
background: var(--teal);
box-shadow: 0 0 0 1px var(--teal);
position: absolute;
left: 79px;
top: 6px;
}
.date-main {
font-size: 0.86rem;
font-weight: 780;
color: var(--ink);
}
.model h3 {
margin: 0 0 7px;
font-size: 1rem;
line-height: 1.25;
}
.org,
.kind {
color: var(--muted);
font-size: 0.9rem;
line-height: 1.4;
}
.desc {
margin: 0 0 10px;
color: #33413d;
font-size: 0.93rem;
line-height: 1.48;
}
.meta {
display: flex;
flex-wrap: wrap;
gap: 6px;
}
.empty {
display: none;
padding: 32px;
text-align: center;
color: var(--muted);
border: 1px dashed var(--line-strong);
border-radius: var(--radius);
background: rgba(255, 253, 248, 0.72);
}
.notes {
margin-top: 22px;
padding: 18px 20px;
border: 1px solid var(--line);
border-radius: var(--radius);
background: rgba(255, 253, 248, 0.78);
color: var(--muted);
font-size: 0.9rem;
line-height: 1.55;
}
@media (max-width: 880px) {
.page { width: min(100vw - 22px, 720px); padding-top: 24px; }
header { grid-template-columns: 1fr; align-items: start; }
.stats { grid-template-columns: repeat(4, minmax(0, 1fr)); }
.toolbar { grid-template-columns: 1fr; position: static; }
.release { grid-template-columns: 112px 1fr; }
.release::before { left: 68px; }
.dot { left: 49px; }
.org, .kind, .desc-wrap { grid-column: 2; }
}
@media (max-width: 560px) {
.stats { grid-template-columns: repeat(2, minmax(0, 1fr)); }
.quarter > summary,
.month > summary {
align-items: flex-start;
flex-direction: column;
}
.release {
grid-template-columns: 1fr;
gap: 8px;
}
.release::before,
.dot { display: none; }
.org, .kind, .desc-wrap { grid-column: auto; }
.date-cell { flex-direction: row; flex-wrap: wrap; }
}
</style>
</head>
<body>
<main class="page">
<header>
<div>
<h1>Model-Releases 2025-2026</h1>
<p class="intro">Eine trainer-taugliche Übersicht der großen öffentlichen KI-Modellreleases vom 1. Januar 2025 bis 2. Juli 2026. Quartale und Monate sind aufklappbar; für Releases ab Januar 2026 sind Tagesangaben und Context-Fenster ergänzt, soweit belastbar auffindbar.</p>
</div>
<section class="stats" aria-label="Kennzahlen">
<div class="stat"><b id="stat-models">0</b><span>Releases</span></div>
<div class="stat"><b id="stat-months">0</b><span>Monate</span></div>
<div class="stat"><b id="stat-context">0</b><span>Context-Angaben</span></div>
<div class="stat"><b>18</b><span>Monate plus heute</span></div>
</section>
</header>
<section class="toolbar" aria-label="Werkzeuge">
<input id="search" type="search" placeholder="Model, Organisation, Region oder Typ filtern">
<button type="button" id="expandAll">Alle öffnen</button>
<button type="button" id="collapseAll">Alle schließen</button>
</section>
<div class="legend" aria-label="Legende">
<span class="pill day">Tag verifiziert oder Quelle nennt konkretes Datum</span>
<span class="pill context">Maximale Context-Size ergänzt</span>
<span class="pill source">Quelle aus Anhang bzw. Primär-/Fachquelle</span>
</div>
<section id="timeline" aria-label="Release-Timeline"></section>
<p id="empty" class="empty">Keine Einträge für diesen Filter.</p>
<aside class="notes">
Hinweis: Die Ausgangsliste ist bewusst breit und nach Modellfamilien konsolidiert. Bei Einträgen der letzten sechs Monate steht „Tag nicht veröffentlicht“, wenn öffentlich keine belastbare Tagesangabe erkennbar war; Context-Größen sind nur eingetragen, wenn sie in den Quellen genannt wurden oder bereits klar in der gelieferten Liste standen.
</aside>
</main>
<script>
const releases = [
{
"month": "2025-01",
"model": "Codestral 25.01",
"org": "Mistral, France",
"type": "code",
"desc": "Faster coding/completion update to Codestral, especially for fill-in-the-middle workflows.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/codestral-2501/"
},
{
"month": "2025-01",
"model": "DeepSeek-R1",
"org": "China",
"type": "reasoning / open-weight",
"desc": "Open reasoning model that helped define the 2025 open-reasoning wave; later updated as R1-0528.",
"source": "DeepSeek",
"url": "https://api-docs.deepseek.com/news/news250120"
},
{
"month": "2025-01",
"model": "Kimi K1.5",
"org": "Moonshot AI, China",
"type": "reasoning / multimodal",
"desc": "Reasoning-focused Kimi update positioned against OpenAI o1 on math, coding, and multimodal reasoning.",
"source": "Kimi / Moonshot history",
"url": "https://www.moonshot.ai/",
"day": "20. Januar 2025",
"context": "nicht angegeben",
"note": "Kimi-Historie nennt K1.5 als Januar-2025-Release"
},
{
"month": "2025-01",
"model": "Luma Ray2",
"org": "US",
"type": "video",
"desc": "Video foundation model focused on realistic motion, coherent action, and production-ready video generation.",
"source": "AWS",
"url": "https://aws.amazon.com/about-aws/whats-new/2025/01/luma-ais-ray2-visual-ai-model-amazon-bedrock/"
},
{
"month": "2025-01",
"model": "Mistral Small 3",
"org": "France / EU",
"type": "small / open",
"desc": "24B latency-optimized Apache-2.0 model, designed for local/private inference and low-latency assistants.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/mistral-small-3/"
},
{
"month": "2025-01",
"model": "OpenAI o3-mini",
"org": "US",
"type": "reasoning / small",
"desc": "Cost-efficient reasoning model for STEM, coding, math, and science, with adjustable reasoning effort in the API.",
"source": "OpenAI",
"url": "https://openai.com/index/openai-o3-mini/"
},
{
"month": "2025-01",
"model": "Qwen2.5-Max",
"org": "Alibaba, China",
"type": "frontier / MoE",
"desc": "Large-scale MoE model trained on more than 20T tokens, available via Alibaba Cloud and Qwen Chat.",
"source": "Qwen",
"url": "https://qwenlm.github.io/blog/qwen2.5-max/"
},
{
"month": "2025-01",
"model": "Qwen2.5-VL",
"org": "Alibaba, China",
"type": "vision-language / open",
"desc": "Vision-language family focused on image, video, document, chart, and UI understanding.",
"source": "Qwen",
"url": "https://qwenlm.github.io/blog/qwen2.5-vl/"
},
{
"month": "2025-02",
"model": "Claude 3.7 Sonnet",
"org": "Anthropic, US",
"type": "frontier / hybrid reasoning",
"desc": "First Claude hybrid reasoning model, able to answer quickly or use extended thinking; strong coding and agentic performance.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-3-7-sonnet"
},
{
"month": "2025-02",
"model": "Command R7B Arabic",
"org": "Cohere, Canada",
"type": "small / regional",
"desc": "Lightweight Command-family model optimized for Arabic and MENA enterprise use cases.",
"source": "Cohere",
"url": "https://cohere.com/blog/command-r7b-arabic"
},
{
"month": "2025-02",
"model": "Gemini 2.0 Flash / Flash-Lite / Pro",
"org": "Google, US",
"type": "multimodal",
"desc": "Gemini 2.0 line: Flash for general multimodal use, Flash-Lite for cost, Pro for stronger reasoning; native tool use and long context.",
"source": "Google",
"url": "https://developers.googleblog.com/en/gemini-2-family-expands/"
},
{
"month": "2025-02",
"model": "Grok 3 / Grok 3 mini",
"org": "xAI, US",
"type": "frontier / reasoning",
"desc": "xAI's Grok 3 beta family with reasoning and DeepSearch, followed by API availability.",
"source": "xAI",
"url": "https://x.ai/news/grok-3"
},
{
"month": "2025-02",
"model": "OpenAI GPT-4.5",
"org": "US",
"type": "frontier / research preview",
"desc": "Large non-reasoning chat model preview focused on knowledge, creativity, and conversational quality.",
"source": "OpenAI",
"url": "https://openai.com/index/introducing-gpt-4-5/"
},
{
"month": "2025-02",
"model": "Phi-4-mini / Phi-4-multimodal",
"org": "Microsoft, US",
"type": "small / open",
"desc": "Compact Phi models: mini for text tasks, multimodal for speech, vision, and text in one 5.6B model.",
"source": "Microsoft",
"url": "https://azure.microsoft.com/en-us/blog/empowering-innovation-the-next-generation-of-the-phi-family/"
},
{
"month": "2025-03",
"model": "Baidu ERNIE 4.5 / ERNIE X1",
"org": "China",
"type": "multimodal / reasoning",
"desc": "ERNIE 4.5 as multimodal foundation model and X1 as reasoning model, initially through Baidu's Qianfan platform.",
"source": "PR Newswire",
"url": "https://www.prnewswire.com/news-releases/baidu-unveils-ernie-4-5-and-reasoning-model-ernie-x1--makes-ernie-bot-free-ahead-of-schedule-302402490.html"
},
{
"month": "2025-03",
"model": "Command A",
"org": "Cohere, Canada",
"type": "enterprise LLM",
"desc": "Enterprise-oriented model for RAG, tool use, multilingual work, and agentic tasks.",
"source": "Cohere",
"url": "https://cohere.com/blog/command-a"
},
{
"month": "2025-03",
"model": "Gemma 3",
"org": "Google, US",
"type": "open / small-to-mid",
"desc": "Open model family in 1B, 4B, 12B, and 27B sizes, with vision-language input, 128K context, and 140+ languages.",
"source": "Google",
"url": "https://developers.googleblog.com/introducing-gemma3/"
},
{
"month": "2025-03",
"model": "Mistral Small 3.1",
"org": "France / EU",
"type": "open / small",
"desc": "Apache-2.0 multimodal 24B-class model with 128K context, fast inference, and improved text/image understanding.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/mistral-small-3-1/"
},
{
"month": "2025-03",
"model": "OLMo 2 32B",
"org": "Ai2, US",
"type": "fully open",
"desc": "Fully open 32B model with data, code, weights, and training details released.",
"source": "Ai2",
"url": "https://allenai.org/blog/olmo2-32b"
},
{
"month": "2025-03",
"model": "OpenAI GPT-4o-transcribe / audio models",
"org": "US",
"type": "speech",
"desc": "New speech-to-text and audio models, improving WER and language recognition over Whisper-era models.",
"source": "OpenAI",
"url": "https://openai.com/index/introducing-our-next-generation-audio-models/"
},
{
"month": "2025-03",
"model": "Runway Gen-4",
"org": "US",
"type": "video",
"desc": "Video model focused on world consistency: keeping characters, objects, style, and scenes coherent across shots.",
"source": "Runway",
"url": "https://runwayml.com/research/introducing-runway-gen-4"
},
{
"month": "2025-04",
"model": "Amazon Nova Premier",
"org": "Amazon, US",
"type": "frontier / enterprise",
"desc": "Most capable Nova model, positioned for complex tasks and as a teacher model for distillation.",
"source": "AWS",
"url": "https://aws.amazon.com/blogs/aws/amazon-nova-premier-our-most-capable-model-for-complex-tasks-and-teacher-for-model-distillation/"
},
{
"month": "2025-04",
"model": "Gemini 2.5 Flash",
"org": "Google, US",
"type": "hybrid reasoning",
"desc": "Cost/speed-focused Gemini 2.5 model with configurable thinking budget.",
"source": "Google",
"url": "https://developers.googleblog.com/en/start-building-with-gemini-25-flash/"
},
{
"month": "2025-04",
"model": "GPT-image-1 / 4o image generation",
"org": "OpenAI, US",
"type": "image",
"desc": "Image generation/editing model released globally via API after ChatGPT image generation rollout.",
"source": "OpenAI",
"url": "https://openai.com/index/image-generation-api/"
},
{
"month": "2025-04",
"model": "Kimi-VL",
"org": "Moonshot AI, China",
"type": "open / vision-language",
"desc": "Efficient open-source MoE vision-language model with long-context understanding and agent capabilities.",
"source": "Hugging Face",
"url": "https://huggingface.co/moonshotai/Kimi-VL-A3B-Instruct",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "April-2025-Release in Kimi-Historie"
},
{
"month": "2025-04",
"model": "Llama 4 Scout / Maverick",
"org": "Meta, US",
"type": "open-weight / multimodal",
"desc": "First Llama 4 open-weight multimodal MoE models; Scout optimized for long context and single-GPU-style deployment.",
"source": "Meta",
"url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/"
},
{
"month": "2025-04",
"model": "OpenAI GPT-4.1 / mini / nano",
"org": "US",
"type": "coding / long-context",
"desc": "API models focused on coding, instruction following, and up to 1M-token context.",
"source": "OpenAI",
"url": "https://openai.com/index/gpt-4-1/"
},
{
"month": "2025-04",
"model": "OpenAI o3 / o4-mini",
"org": "US",
"type": "reasoning",
"desc": "Stronger reasoning models, with o4-mini positioned for faster/cost-efficient reasoning; o3-pro followed for reliability.",
"source": "OpenAI",
"url": "https://openai.com/index/introducing-o3-and-o4-mini/"
},
{
"month": "2025-04",
"model": "Phi-4-reasoning / reasoning-plus",
"org": "Microsoft, US",
"type": "small / open-weight reasoning",
"desc": "14B open-weight reasoning model trained from Phi-4, aimed at complex reasoning with detailed reasoning chains.",
"source": "Microsoft",
"url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/"
},
{
"month": "2025-04",
"model": "Qwen3 family",
"org": "Alibaba, China",
"type": "open / dense+MoE",
"desc": "Large open model family with dense and MoE variants, reasoning/coding/STEM focus, multilingual support.",
"source": "Qwen",
"url": "https://qwenlm.github.io/blog/qwen3/"
},
{
"month": "2025-05",
"model": "Claude Opus 4 / Sonnet 4",
"org": "Anthropic, US",
"type": "frontier / hybrid reasoning",
"desc": "Claude 4 family for coding, agents, extended thinking, and long-running workflows.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-4"
},
{
"month": "2025-05",
"model": "DeepSeek-R1-0528",
"org": "DeepSeek, China",
"type": "reasoning / open-weight refresh",
"desc": "Updated R1 reasoning checkpoint with improved reasoning and tool-oriented behavior after the January R1 release.",
"source": "DeepSeek Docs",
"url": "https://api-docs.deepseek.com/news/news250528",
"day": "28. Mai 2025",
"context": "nicht angegeben",
"note": "Datum aus DeepSeek-News-Slug"
},
{
"month": "2025-05",
"model": "Devstral Small",
"org": "Mistral, France",
"type": "code / open",
"desc": "Apache-2.0 coding-agent model for autonomous software development workflows.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/devstral/"
},
{
"month": "2025-05",
"model": "Falcon Arabic / Falcon-H1",
"org": "UAE",
"type": "regional / open",
"desc": "Falcon Arabic for Arabic-language AI and Falcon-H1 as efficient hybrid architecture model family.",
"source": "TII",
"url": "https://www.tii.ae/news/middle-easts-leading-ai-powerhouse-tii-launches-two-new-ai-models-falcon-arabic-first-arabic"
},
{
"month": "2025-05",
"model": "Mistral Medium 3",
"org": "France / EU",
"type": "frontier-ish / efficient",
"desc": "Closed/API model positioned as high-performance at lower cost, near frontier class for enterprise tasks.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/mistral-medium-3"
},
{
"month": "2025-05",
"model": "Sarvam-M",
"org": "India",
"type": "regional / reasoning",
"desc": "Indian-language-focused LLM built on Mistral Small and post-trained for Indian languages, math, and programming.",
"source": "Sarvam AI",
"url": "https://www.sarvam.ai/blogs/sarvam-m"
},
{
"month": "2025-05",
"model": "Veo 3 / Imagen 4 / Lyria 2",
"org": "Google, US",
"type": "video / image / music",
"desc": "Google I/O 2025 media models: Veo 3 for video with audio, Imagen 4 for image generation, Lyria 2 for music tools.",
"source": "Google",
"url": "https://blog.google/innovation-and-ai/products/generative-media-models-io-2025/"
},
{
"month": "2025-06",
"model": "Baidu ERNIE 4.5 open family",
"org": "China",
"type": "open / multimodal",
"desc": "Ten-model open-source ERNIE 4.5 family, including large MoE and small dense variants.",
"source": "Baidu",
"url": "https://ernie.baidu.com/blog/posts/ernie4.5/"
},
{
"month": "2025-06",
"model": "HyperCLOVA X THINK",
"org": "NAVER, Korea",
"type": "reasoning / Korean-English",
"desc": "Korean-English reasoning model with 128K context and Korean synthetic data focus.",
"source": "NAVER",
"url": "https://clova.ai/en/tech-blog/6203-2"
},
{
"month": "2025-06",
"model": "Kimi-Dev",
"org": "Moonshot AI, China",
"type": "code / open",
"desc": "72B coding-focused model based on Qwen2.5-72B, released for software-engineering benchmarks and agents.",
"source": "Hugging Face",
"url": "https://huggingface.co/moonshotai/Kimi-Dev-72B",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Juni-2025-Release in Kimi-Historie"
},
{
"month": "2025-06",
"model": "Kimi-VL-Thinking",
"org": "Moonshot AI, China",
"type": "vision-language / reasoning",
"desc": "Thinking variant of Kimi-VL for multimodal reasoning with longer decoding.",
"source": "Hugging Face",
"url": "https://huggingface.co/moonshotai/Kimi-VL-A3B-Thinking-2506",
"day": "21. Juni 2025",
"context": "nicht angegeben",
"note": "Hugging-Face-Artikel/Modellseite im Juni 2025"
},
{
"month": "2025-06",
"model": "Magistral Small / Medium",
"org": "Mistral, France",
"type": "reasoning",
"desc": "Mistral's first reasoning models; Small is 24B open, Medium is stronger enterprise model.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/magistral/"
},
{
"month": "2025-06",
"model": "MiniMax-M1",
"org": "China",
"type": "open-weight reasoning",
"desc": "456B total / 45.9B active hybrid-attention reasoning model with long-context and tool/coding focus.",
"source": "MiniMax",
"url": "https://www.minimax.io/news/minimaxm1"
},
{
"month": "2025-06",
"model": "OpenAI o3-pro",
"org": "OpenAI, US",
"type": "reasoning / pro",
"desc": "Longer-thinking o3 variant for more reliable answers, made available in ChatGPT Pro and the API.",
"source": "OpenAI",
"url": "https://openai.com/index/introducing-o3-and-o4-mini/",
"day": "10. Juni 2025",
"context": "nicht angegeben",
"note": "OpenAI nennt o3-pro-Verfuegbarkeit am 10. Juni 2025"
},
{
"month": "2025-06",
"model": "Qwen3 Embedding / Qwen3 Reranker",
"org": "Alibaba, China",
"type": "embedding / reranking",
"desc": "Open embedding and reranking family in 0.6B, 4B, and 8B sizes for multilingual retrieval and RAG pipelines.",
"source": "Qwen",
"url": "https://qwenlm.github.io/blog/qwen3-embedding/",
"day": "5. Juni 2025",
"context": "bis 32K",
"note": "Qwen-Blog nennt 5. Juni 2025; technische Reports nennen lange Eingaben"
},
{
"month": "2025-07",
"model": "EXAONE 4.0",
"org": "LG AI Research, Korea",
"type": "open-weight / reasoning",
"desc": "32B and 1.2B hybrid reasoning/non-reasoning models, focused on Korean, English, Spanish, tool use, and agents.",
"source": "GitHub",
"url": "https://github.com/LG-AI-EXAONE/EXAONE-4.0"
},
{
"month": "2025-07",
"model": "Grok 4",
"org": "xAI, US",
"type": "frontier / reasoning",
"desc": "Frontier multimodal reasoning model with native tool use and live search integration.",
"source": "xAI",
"url": "https://x.ai/news/grok-4",
"day": "9. Juli 2025",
"context": "bis 256K",
"note": "xAI nennt 256K Context Window"
},
{
"month": "2025-07",
"model": "Kimi K2",
"org": "Moonshot, China",
"type": "open / MoE / code-agent",
"desc": "1T-parameter MoE, 32B active, open-source model focused on agentic coding and tool use.",
"source": "arXiv",
"url": "https://arxiv.org/abs/2507.20534"
},
{
"month": "2025-07",
"model": "Qwen3-235B / Qwen-MT",
"org": "Alibaba, China",
"type": "reasoning / translation",
"desc": "Additional Qwen3 releases for complex reasoning, coding, and machine translation.",
"source": "Alibaba Cloud",
"url": "https://www.alibabacloud.com/blog/alibaba-unveils-new-qwen3-models-for-coding-complexing-reasoning-and-machine-translation_602415"
},
{
"month": "2025-07",
"model": "Qwen3-Coder",
"org": "Alibaba, China",
"type": "code / agentic",
"desc": "Qwen's most agentic coding model at launch, aimed at code generation, debugging, and large-codebase workflows.",
"source": "Alibaba Cloud",
"url": "https://www.alibabacloud.com/blog/alibaba-unveils-cutting-edge-ai-coding-model-qwen3-coder_602399"
},
{
"month": "2025-07",
"model": "Solar Pro 2",
"org": "Upstage, Korea",
"type": "frontier-ish / regional",
"desc": "Korean-developed LLM recognized by Artificial Analysis as a leading frontier-level model.",
"source": "Financial Times",
"url": "https://www.ft.com/content/549a34f1-5eb5-43bd-8400-2171e6e32022"
},
{
"month": "2025-07",
"model": "Voxtral Small / Mini",
"org": "Mistral, France",
"type": "speech / open",
"desc": "Apache-2.0 speech-understanding models, including voice-to-function-calling workflows.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/voxtral/"
},
{
"month": "2025-08",
"model": "Claude Opus 4.1",
"org": "Anthropic, US",
"type": "frontier refresh",
"desc": "Upgrade to Opus 4 for agentic tasks, coding, and reasoning.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-opus-4-1"
},
{
"month": "2025-08",
"model": "Gemini 2.5 Flash Image / nano-banana",
"org": "Google, US",
"type": "image",
"desc": "Image generation/editing model with local edits, multi-image fusion, and Gemini world knowledge.",
"source": "Google",
"url": "https://developers.googleblog.com/en/introducing-gemini-2-5-flash-image/"
},
{
"month": "2025-08",
"model": "GLM-4.5 / GLM-4.5-Air",
"org": "Z.ai / Zhipu, China",
"type": "open / agentic / MoE",
"desc": "Open-source MoE model family for agents and reasoning, with hybrid thinking/direct modes.",
"source": "Reuters",
"url": "https://www.reuters.com/technology/chinas-ai-startup-zhipu-releases-open-source-model-glm-45-2025-07-28/"
},
{
"month": "2025-08",
"model": "GLM-4.5V",
"org": "Z.ai / Zhipu, China",
"type": "vision reasoning",
"desc": "100B-scale open-source vision reasoning model for video, grounding, and GUI agents.",
"source": "Z.AI",
"url": "https://docs.z.ai/release-notes/new-released"
},
{
"month": "2025-08",
"model": "OpenAI GPT-5",
"org": "US",
"type": "frontier",
"desc": "Default ChatGPT model release, positioned for stronger coding, reasoning, writing, and task routing.",
"source": "OpenAI",
"url": "https://openai.com/index/introducing-gpt-5/"
},
{
"month": "2025-08",
"model": "Qwen3-Coder-Next",
"org": "Alibaba, China",
"type": "code / efficient hybrid",
"desc": "Smaller hybrid-model line focused on agentic coding efficiency after the Qwen3-Coder flagship.",
"source": "Qwen",
"url": "https://qwen.ai/blog?id=qwen3-coder-next",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Qwen-Blog als Zwischenrelease der Coder-Linie"
},
{
"month": "2025-09",
"model": "Apertus 8B / 70B",
"org": "Switzerland / EU",
"type": "fully open / multilingual",
"desc": "Swiss fully open foundation model with weights, data, documentation, and multilingual coverage.",
"source": "ETH Zürich",
"url": "https://ethz.ch/en/news-and-events/eth-news/news/2025/09/press-release-apertus-a-fully-open-transparent-multilingual-language-model.html"
},
{
"month": "2025-09",
"model": "Claude Sonnet 4.5",
"org": "Anthropic, US",
"type": "frontier / agents",
"desc": "Strong coding and agentic model, with long-running task focus and broad domain reasoning improvements.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-sonnet-4-5"
},
{
"month": "2025-09",
"model": "DeepSeek-V3.1 / V3.1-Terminus",
"org": "China",
"type": "open / general+reasoning",
"desc": "DeepSeek model refreshes bridging chat and reasoning variants, later leading to V3.2.",
"source": "DeepSeek",
"url": "https://api-docs.deepseek.com/news/news250821"
},
{
"month": "2025-09",
"model": "Grok 4 Fast",
"org": "xAI, US",
"type": "fast reasoning / search",
"desc": "Cost-efficient Grok 4 variant with unified reasoning/non-reasoning modes and web/X search capability.",
"source": "xAI",
"url": "https://x.ai/news/grok-4-fast",
"day": "Tag nicht veröffentlicht",
"context": "bis 2M",
"note": "xAI nennt 2M Token Context Window"
},
{
"month": "2025-09",
"model": "Kimi K2-Instruct-0905",
"org": "Moonshot AI, China",
"type": "open / MoE / code-agent refresh",
"desc": "September K2 update with stronger agentic coding and frontend-coding behavior.",
"source": "Hugging Face",
"url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905",
"day": "5. September 2025",
"context": "bis 256K",
"note": "Hugging Face nennt 0905 und erweiterten Context von 128K auf 256K"
},
{
"month": "2025-09",
"model": "OpenAI Sora 2",
"org": "US",
"type": "video+audio",
"desc": "Flagship video/audio generation model with improved physics, realism, synchronized audio, and steerability.",
"source": "OpenAI",
"url": "https://openai.com/index/sora-2/"
},
{
"month": "2025-09",
"model": "Qwen3-Max",
"org": "Alibaba, China",
"type": "frontier / very large",
"desc": "Alibaba's largest and most capable Qwen model at launch, above 1T parameters per Reuters.",
"source": "Qwen",
"url": "https://qwen.ai/blog?id=qwen3-max"
},
{
"month": "2025-09",
"model": "Qwen3-Next",
"org": "Alibaba, China",
"type": "efficient MoE",
"desc": "80B-A3B variants with Instruct and Thinking modes for efficient deployment.",
"source": "Qwen",
"url": "https://qwen.ai/"
},
{
"month": "2025-09",
"model": "Qwen3-VL / Qwen3-Omni",
"org": "Alibaba, China",
"type": "vision / omni-modal",
"desc": "Qwen3-VL for vision-language action; Qwen3-Omni for text, images, audio, video, and real-time streaming.",
"source": "Qwen",
"url": "https://qwen.ai/"
},
{
"month": "2025-10",
"model": "Claude Haiku 4.5",
"org": "Anthropic, US",
"type": "small / fast",
"desc": "Small Claude model with Sonnet-4-class coding performance at lower cost and higher speed.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-haiku-4-5"
},
{
"month": "2025-10",
"model": "Gemini 2.5 Computer Use",
"org": "Google, US",
"type": "agent / computer use",
"desc": "Specialized model for controlling browsers/interfaces and performing computer-use tasks.",
"source": "Google",
"url": "https://ai.google.dev/gemini-api/docs/changelog"
},
{
"month": "2025-10",
"model": "GLM-4.6",
"org": "Z.ai / Zhipu, China",
"type": "agentic / reasoning / coding",
"desc": "GLM-4 series upgrade for advanced agentic, reasoning, and coding capabilities.",
"source": "Z.ai",
"url": "https://z.ai/blog/glm-4.6",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Z.ai-Blog als Zwischenrelease zwischen GLM-4.5 und GLM-4.7"
},
{
"month": "2025-10",
"model": "IBM Granite 4.0",
"org": "IBM, US",
"type": "open / enterprise",
"desc": "Hybrid Mamba/Transformer architecture for efficient enterprise language models, lower memory, faster inference.",
"source": "IBM",
"url": "https://www.ibm.com/new/announcements/ibm-granite-4-0-hyper-efficient-high-performance-hybrid-models"
},
{
"month": "2025-10",
"model": "Kimi Linear 48B-A3B",
"org": "Moonshot AI, China",
"type": "open / efficient long-context",
"desc": "Kimi Delta Attention model reducing KV cache needs and improving long-context decoding throughput.",
"source": "Hugging Face",
"url": "https://huggingface.co/moonshotai/Kimi-Linear-48B-A3B-Instruct",
"day": "30. Oktober 2025",
"context": "bis 1M",
"note": "Kimi Linear Paper/Modelkarte nennt 1M-Kontexte"
},
{
"month": "2025-10",
"model": "MiniMax-M2",
"org": "China",
"type": "open / code-agent",
"desc": "Compact open MoE for coding and agent workflows, 230B total / 10B active.",
"source": "MiniMax",
"url": "https://www.minimax.io/news/minimax-m2"
},
{
"month": "2025-11",
"model": "Claude Opus 4.5",
"org": "Anthropic, US",
"type": "frontier",
"desc": "Model for coding, agents, computer use, deep research, slides, and spreadsheets.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-opus-4-5"
},
{
"month": "2025-11",
"model": "FLUX.2",
"org": "Black Forest Labs, Germany/US",
"type": "image",
"desc": "Frontier visual intelligence model from BFL, positioned as its most capable FLUX release.",
"source": "Black Forest Labs",
"url": "https://blackforestlabs.ai/announcements/"
},
{
"month": "2025-11",
"model": "Gemini 3 Pro / Gemini 3 Pro Image",
"org": "Google, US",
"type": "frontier / multimodal",
"desc": "Gemini 3 as Google's most intelligent model at release, with reasoning, multimodal understanding, and Search integration from day one.",
"source": "Google",
"url": "https://blog.google/products-and-platforms/products/gemini/gemini-3/"
},
{
"month": "2025-11",
"model": "Kimi K2 Thinking",
"org": "Moonshot AI, China",
"type": "open / reasoning / agentic",
"desc": "Open-source thinking model extending K2 with long-horizon reasoning, tool orchestration, and native INT4 quantization.",
"source": "Hugging Face",
"url": "https://huggingface.co/moonshotai/Kimi-K2-Thinking",
"day": "6. November 2025",
"context": "bis 256K",
"note": "Modelkarte nennt 256K Context Length"
},
{
"month": "2025-11",
"model": "OLMo 3",
"org": "Ai2, US",
"type": "fully open",
"desc": "7B and 32B fully open model family covering base, instruct, and thinking variants with full model-flow release.",
"source": "Ai2",
"url": "https://allenai.org/blog/olmo3"
},
{
"month": "2025-11",
"model": "OpenAI GPT-5.1",
"org": "OpenAI, US",
"type": "frontier / chat+reasoning refresh",
"desc": "GPT-5 refresh with Instant and Thinking variants, stronger instruction-following and conversational behavior.",
"source": "OpenAI",
"url": "https://openai.com/index/gpt-5-1/",
"day": "13. November 2025",
"context": "nicht angegeben",
"note": "API-Ankuendigung am 13. November 2025"
},
{
"month": "2025-12",
"model": "Amazon Nova 2 Lite / Nova 2 Sonic",
"org": "Amazon, US",
"type": "reasoning / speech",
"desc": "Nova 2 Lite for fast cost-effective reasoning; Nova 2 Sonic for speech-to-speech conversational AI.",
"source": "AWS",
"url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-lite-a-fast-cost-effective-reasoning-model/"
},
{
"month": "2025-12",
"model": "DeepSeek-V3.2",
"org": "China",
"type": "open / reasoning-agent",
"desc": "Efficiency-focused model with strong reasoning and agent performance; Speciale variant scored highly in competition-style reasoning.",
"source": "DeepSeek",
"url": "https://api-docs.deepseek.com/news/news251201"
},
{
"month": "2025-12",
"model": "Devstral 2 / Devstral Small 2",
"org": "Mistral, France",
"type": "code / open",
"desc": "Next-generation coding model family, 123B and 24B, for code agents and autonomous development.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/devstral-2-vibe-cli/"
},
{
"month": "2025-12",
"model": "EuroLLM-22B",
"org": "Europe",
"type": "open / multilingual",
"desc": "European open LLM release for researchers, startups, developers, and public institutions.",
"source": "IT.pt",
"url": "https://www.it.pt/News/NewsPost/5203"
},
{
"month": "2025-12",
"model": "Gemini 3 Flash",
"org": "Google, US",
"type": "fast frontier",
"desc": "Efficient Gemini 3-family model for CLI/coding and high-frequency workloads.",
"source": "Google",
"url": "https://developers.googleblog.com/gemini-3-flash-is-now-available-in-gemini-cli/"
},
{
"month": "2025-12",
"model": "MiniLingua",
"org": "Europe",
"type": "small / multilingual",
"desc": "1B open-source LLM for 13 European languages.",
"source": "arXiv",
"url": "https://arxiv.org/abs/2512.13298"
},
{
"month": "2025-12",
"model": "Mistral 3 / Mistral Large 3 / Ministral 3",
"org": "France / EU",
"type": "open / multimodal",
"desc": "Apache-2.0 family: Large 3 MoE plus 14B, 8B, 3B dense small models, with vision and 256K context.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/mistral-3/"
},
{
"month": "2025-12",
"model": "Mistral OCR 3",
"org": "France / EU",
"type": "document AI",
"desc": "Document extraction model with markdown/HTML table reconstruction, optimized for low cost.",
"source": "Mistral AI",
"url": "https://mistral.ai/news/mistral-ocr-3/"
},
{
"month": "2025-12",
"model": "NVIDIA Nemotron 3 Nano",
"org": "NVIDIA, US",
"type": "open / agentic",
"desc": "Open model family for agentic AI; Nano released first, Super/Ultra planned for 2026.",
"source": "NVIDIA",
"url": "https://developer.nvidia.com/blog/inside-nvidia-nemotron-3-techniques-tools-and-data-that-make-it-efficient-and-accurate/"
},
{
"month": "2025-12",
"model": "OpenAI GPT-5.2 / GPT-5.2-Codex",
"org": "OpenAI, US",
"type": "frontier / agentic / code",
"desc": "GPT-5.2 family for professional work, long-context understanding, agentic tool-calling, vision, and Codex surfaces.",
"source": "OpenAI",
"url": "https://openai.com/index/introducing-gpt-5-2/",
"day": "11. Dezember 2025",
"context": "nicht angegeben",
"note": "OpenAI nennt GPT-5.2-Rollout und separate Codex-Verfuegbarkeit"
},
{
"month": "2025-12",
"model": "Runway Gen-4.5",
"org": "US",
"type": "video",
"desc": "Video model focused on visual fidelity, physical accuracy, motion quality, and controllable generation.",
"source": "Runway",
"url": "https://runwayml.com/research/introducing-runway-gen-4.5"
},
{
"month": "2026-01",
"model": "AI21 Jamba2 / Jamba mini 2",
"org": "Israel",
"type": "hybrid LLM",
"desc": "Jamba2 generation following 2025 Jamba updates, with current API aliases in 2026.",
"source": "AI21",
"url": "https://www.ai21.com/jamba/",
"day": "Tag nicht veröffentlicht",
"context": "bis 256K",
"note": "Context je nach Jamba-Variante/API"
},
{
"month": "2026-01",
"model": "GLM-4.7",
"org": "Z.ai / Zhipu, China",
"type": "coding / agentic",
"desc": "Coding-focused GLM update for coding agents such as Claude Code, Kilo Code, Roo Code, and Cline.",
"source": "Z.ai",
"url": "https://z.ai/blog/glm-4.7",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Z.ai-Blog als Zwischenrelease vor GLM-5"
},
{
"month": "2026-01",
"model": "HyperCLOVA X 32B Think",
"org": "NAVER, Korea",
"type": "open / VLM reasoning",
"desc": "Vision-language model focused on Korean cultural/linguistic reasoning and agentic behavior.",
"source": "arXiv",
"url": "https://arxiv.org/abs/2601.03286",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "arXiv-Monat belastbar, Tagesangabe nicht in der Liste"
},
{
"month": "2026-01",
"model": "Kimi K2.5",
"org": "Moonshot AI, China",
"type": "multimodal / agentic",
"desc": "Visual-agentic Kimi generation with instant and thinking modes, multimodal understanding, and agentic workflows.",
"source": "Hugging Face",
"url": "https://huggingface.co/moonshotai/Kimi-K2.5",
"day": "Tag nicht veröffentlicht",
"context": "bis 256K",
"note": "Kimi K2.5 Experimente mit 256K Context Length"
},
{
"month": "2026-01",
"model": "Qwen3-VL-Embedding / Qwen3-VL-Reranker",
"org": "Alibaba, China",
"type": "multimodal embedding / reranking",
"desc": "Multimodal retrieval and ranking family built on Qwen3-VL for text, images, document images, and video.",
"source": "arXiv",
"url": "https://arxiv.org/abs/2601.04720",
"day": "8. Januar 2026",
"context": "bis 32K",
"note": "arXiv-Erstversion vom 8. Januar 2026"
},
{
"month": "2026-02",
"model": "Claude Sonnet 4.6",
"org": "Anthropic, US",
"type": "frontier / efficient",
"desc": "Upgrade bringing stronger coding, computer use, document understanding, and instruction-following at Sonnet cost.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-sonnet-4-6",
"day": "Tag nicht veröffentlicht",
"context": "bis 200K",
"note": "Claude-Plattform nennt 200K Context-Fenster fuer Sonnet-Familie"
},
{
"month": "2026-02",
"model": "Gemini 3.1 Pro / 3.1 Flash Image",
"org": "Google, US",
"type": "frontier / image",
"desc": "3.1 Pro improved reasoning for complex tasks; 3.1 Flash Image followed as image-generation/editing update.",
"source": "Google DeepMind",
"url": "https://deepmind.google/blog/gemini-3-1-pro-a-smarter-model-for-your-most-complex-tasks/",
"day": "Tag nicht veröffentlicht",
"context": "bis 1M",
"note": "Gemini-Pro-Familie mit langem Context; Bildmodell separat betrachten"
},
{
"month": "2026-02",
"model": "GLM-5",
"org": "Z.ai / Zhipu, China",
"type": "frontier / agentic",
"desc": "Flagship model aimed at complex systems engineering and long-horizon agentic tasks.",
"source": "Z.ai",
"url": "https://z.ai/blog/glm-5",
"day": "Tag nicht veröffentlicht",
"context": "bis 1M",
"note": "GLM-5-Linie fuer lange Agentic-Kontexte"
},
{
"month": "2026-02",
"model": "OpenAI GPT-5.3-Codex",
"org": "US",
"type": "code-agent",
"desc": "OpenAI release notes describe it as a highly capable agentic coding model combining Codex and GPT-5 stacks.",
"source": "OpenAI Help",
"url": "https://help.openai.com/en/articles/9624314-model-release-notes",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Release-Notes ohne stabile Context-Angabe im Anhang"
},
{
"month": "2026-02",
"model": "Qwen-Image-2.0",
"org": "Alibaba, China",
"type": "image",
"desc": "Foundation image generation model aimed at professional infographics and high-quality visual generation.",
"source": "Qwen",
"url": "https://qwen.ai/blog?id=qwen-image-2.0",
"day": "Tag nicht veröffentlicht",
"context": "nicht relevant",
"note": "Bildmodell statt Text-Context-Fenster"
},
{
"month": "2026-02",
"model": "Qwen3.5",
"org": "Alibaba, China",
"type": "agentic / multimodal",
"desc": "Reuters describes Qwen3.5 as an agentic-era model with cost-efficiency and visual-agentic capabilities.",
"source": "Reuters",
"url": "https://www.reuters.com/world/china/alibaba-unveils-new-qwen35-model-agentic-ai-era-2026-02-16/",
"day": "16. Februar 2026",
"context": "nicht angegeben",
"note": "Reuters-Datum aus Quellen-URL"
},
{
"month": "2026-02",
"model": "Sarvam 30B / Sarvam 105B",
"org": "India",
"type": "open / regional",
"desc": "Indian-language-focused open models, including 105B MoE, for 22 Indian languages and regional context.",
"source": "Wikipedia",
"url": "https://en.wikipedia.org/wiki/Sarvam_AI",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Sekundaerquelle; genaue Release-Seite nicht belastbar auffindbar"
},
{
"month": "2026-03",
"model": "NVIDIA Nemotron 3 Super",
"org": "NVIDIA, US",
"type": "open / agentic",
"desc": "Open hybrid Mamba-Transformer MoE model for agentic reasoning.",
"source": "NVIDIA",
"url": "https://developer.nvidia.com/blog/introducing-nemotron-3-super-an-open-hybrid-mamba-transformer-moe-for-agentic-reasoning/",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Context nicht eindeutig auffindbar"
},
{
"month": "2026-03",
"model": "OLMo Hybrid",
"org": "Ai2, US",
"type": "fully open / hybrid",
"desc": "7B fully open hybrid Transformer/RNN family with improved data efficiency.",
"source": "Ai2",
"url": "https://allenai.org/blog/olmohybrid",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Context nicht in Ausgangsliste genannt"
},
{
"month": "2026-03",
"model": "Phi-4-reasoning-vision-15B",
"org": "Microsoft, US",
"type": "open-weight / multimodal reasoning",
"desc": "Compact 15B vision-language reasoning model for math, science, documents, receipts, and UI grounding.",
"source": "Microsoft",
"url": "https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Blog/Report ohne klare Max-Context-Angabe im Anhang"
},
{
"month": "2026-03",
"model": "Qwen3.6-Plus / Qwen3.6-Max-Preview",
"org": "Alibaba, China",
"type": "agentic / frontier refresh",
"desc": "Qwen3.6 proprietary-line update aimed at real-world agents, including Plus and Max-Preview variants.",
"source": "Qwen",
"url": "https://qwen.ai/blog?id=qwen3.6",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Folgt laut Qwen auf Qwen3.5 im Februar"
},
{
"month": "2026-04",
"model": "Claude Opus 4.7",
"org": "Anthropic, US",
"type": "frontier refresh",
"desc": "Upgrade for complex multi-step workflows, tool reliability, and computer/vision-heavy tasks.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-opus-4-7",
"day": "Tag nicht veröffentlicht",
"context": "bis 200K",
"note": "Claude-Plattform nennt 200K Context-Fenster fuer Opus-Familie"
},
{
"month": "2026-04",
"model": "DeepSeek-V4 Pro / Flash Preview",
"org": "China",
"type": "open / long-context",
"desc": "Open V4 preview with 1M context; Pro and Flash variants.",
"source": "DeepSeek",
"url": "https://api-docs.deepseek.com/news/news260424",
"day": "24. April 2026",
"context": "bis 1M",
"note": "Datum aus DeepSeek-News-URL; Context aus Ausgangsliste"
},
{
"month": "2026-04",
"model": "EXAONE 4.5",
"org": "LG AI Research, Korea",
"type": "open-weight / VLM",
"desc": "LG's first open-weight vision-language model, extending EXAONE with 256K context and document-centric multimodal training.",
"source": "arXiv",
"url": "https://arxiv.org/abs/2604.08644",
"day": "Tag nicht veröffentlicht",
"context": "bis 256K",
"note": "Context aus Ausgangsliste"
},
{
"month": "2026-04",
"model": "Kimi K2.6",
"org": "Moonshot AI, China",
"type": "multimodal / coding / agentic",
"desc": "Natively multimodal Kimi model with stronger coding capabilities and agent performance.",
"source": "Moonshot AI",
"url": "https://www.moonshot.ai/",
"day": "20. April 2026",
"context": "nicht angegeben",
"note": "Moonshot-Startseite listet 2026-04-20 Kimi K2.6"
},
{
"month": "2026-04",
"model": "OpenAI GPT Image 2 / ChatGPT Images 2.0",
"org": "OpenAI, US",
"type": "image",
"desc": "Image model update with improved text rendering, multilingual support, and visual control.",
"source": "OpenAI",
"url": "https://openai.com/index/introducing-chatgpt-images-2-0/",
"day": "Tag nicht veröffentlicht",
"context": "nicht relevant",
"note": "Bildmodell statt Text-Context-Fenster"
},
{
"month": "2026-04",
"model": "Qwen3.6-27B",
"org": "Alibaba, China",
"type": "coding / small-mid",
"desc": "Dense 27B coding and agentic model positioned for coding assistants.",
"source": "Alibaba Cloud",
"url": "https://www.alibabacloud.com/blog/qwen3-6-27b-flagship-level-coding-in-a-27b-dense-model_603063",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Context nicht in Ausgangsliste genannt"
},
{
"month": "2026-05",
"model": "Gemini 3.5 Flash",
"org": "Google, US",
"type": "frontier / agentic",
"desc": "Gemini 3.5 Flash released at I/O 2026 for frontier agent and coding performance with Flash speed/cost profile.",
"source": "Google DeepMind",
"url": "https://deepmind.google/blog/gemini-3-5-frontier-intelligence-with-action/",
"day": "Tag nicht veröffentlicht",
"context": "bis 1M",
"note": "Gemini-Flash-Familie mit langem Context; genaue Variante pruefen"
},
{
"month": "2026-05",
"model": "GLM-5.1",
"org": "Z.ai / Zhipu, China",
"type": "open / long-horizon agentic",
"desc": "Open-source GLM-5 series update under MIT license, focused on long-horizon tasks and coding-agent serving.",
"source": "Z.ai",
"url": "https://z.ai/blog/glm-5.1",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Z.ai-Blog als Zwischenrelease zwischen GLM-5 und GLM-5.2"
},
{
"month": "2026-05",
"model": "MiniMax-M2.5 / MiniMax-M3",
"org": "China",
"type": "open / code-agent / multimodal",
"desc": "M2.5 improves coding/agentic benchmarks; M3 introduces production-grade multimodal sparse attention and 1M context.",
"source": "Hugging Face",
"url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.5",
"day": "Tag nicht veröffentlicht",
"context": "bis 1M",
"note": "Context aus Ausgangsliste fuer M3"
},
{
"month": "2026-05",
"model": "OpenAI GPT-5.5 / GPT-5.5 Pro",
"org": "US",
"type": "frontier",
"desc": "OpenAI's April 2026 release notes position GPT-5.5 as a stronger model for computer work, cybersecurity safeguards, and professional tasks.",
"source": "OpenAI",
"url": "https://openai.com/index/introducing-gpt-5-5/",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Context nicht in Ausgangsliste genannt"
},
{
"month": "2026-05",
"model": "Qwen3.6-35B-A3B",
"org": "Alibaba, China",
"type": "open / agentic coding",
"desc": "Open Qwen3.6 MoE coding/agentic model with a 35B-A3B profile for efficient deployment.",
"source": "Qwen",
"url": "https://qwen.ai/blog?id=qwen3.6-35b-a3b",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Qwen-Blog als open Zwischenrelease der 3.6-Linie"
},
{
"month": "2026-06",
"model": "Claude Fable 5 / Mythos 5",
"org": "Anthropic, US",
"type": "frontier / restricted",
"desc": "Fable 5 is broadly available; Mythos 5 is more restricted/invitation-oriented and tied to Project Glasswing/cybersecurity contexts.",
"source": "Claude Docs",
"url": "https://platform.claude.com/docs/en/about-claude/models/overview",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Modelluebersicht ohne belastbare Release-Tagesangabe im Anhang"
},
{
"month": "2026-06",
"model": "Claude Sonnet 5",
"org": "Anthropic, US",
"type": "frontier / agentic",
"desc": "Upgrade to Sonnet 4.6 for coding, agents, and professional work at scale.",
"source": "Anthropic",
"url": "https://www.anthropic.com/news/claude-sonnet-5",
"day": "Tag nicht veröffentlicht",
"context": "bis 200K",
"note": "Claude-Plattform nennt 200K Context-Fenster fuer Sonnet-Familie"
},
{
"month": "2026-06",
"model": "Gemini 3.5 Audio / Computer Use",
"org": "Google, US",
"type": "audio / agent",
"desc": "Gemini 3.5 Audio model card and built-in computer-use capability in Gemini 3.5 Flash.",
"source": "Google DeepMind",
"url": "https://deepmind.google/models/model-cards/gemini-3-5-audio/",
"day": "Tag nicht veröffentlicht",
"context": "nicht relevant",
"note": "Audio/Computer-Use; Text-Context je nach Basismodell"
},
{
"month": "2026-06",
"model": "GLM-5.2",
"org": "Z.ai / Zhipu, China",
"type": "open / frontier-ish",
"desc": "Latest GLM release for long-horizon tasks, coding, agents, and up to 1M-token context.",
"source": "Z.ai",
"url": "https://z.ai/blog/glm-5.2",
"day": "Tag nicht veröffentlicht",
"context": "bis 1M",
"note": "Context aus Ausgangsliste"
},
{
"month": "2026-06",
"model": "NVIDIA Nemotron 3 Ultra",
"org": "NVIDIA, US",
"type": "open / agentic",
"desc": "550B total / 55B active open model with weights, data, and recipes for long-running agents.",
"source": "NVIDIA",
"url": "https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Context nicht in Ausgangsliste genannt"
},
{
"month": "2026-06",
"model": "OpenAI GPT-5.6 Sol / Terra / Luna",
"org": "US",
"type": "preview / frontier",
"desc": "Limited preview: Sol flagship, Terra lower-cost, Luna fast/cost-efficient; family targets software engineering, professional work, science, and cybersecurity.",
"source": "OpenAI",
"url": "https://openai.com/index/previewing-gpt-5-6-sol/",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Preview ohne klare Context-Angabe in Ausgangsliste"
},
{
"month": "2026-06",
"model": "Qwen3.7-Max",
"org": "Alibaba, China",
"type": "agentic frontier",
"desc": "Latest Qwen proprietary model for the agent era, focused on writing/debugging code and automating complex work.",
"source": "Qwen",
"url": "https://qwen.ai/blog?id=qwen3.7",
"day": "Tag nicht veröffentlicht",
"context": "nicht angegeben",
"note": "Qwen3.7-Blog als aktueller Agent-Frontier-Release"
},
{
"month": "2026-07",
"model": "Amália",
"org": "Portugal / EU",
"type": "open / sovereign",
"desc": "Portugal's first open-source foundation model, released with dataset/source code for public-use national AI infrastructure.",
"source": "Reuters",
"url": "https://www.reuters.com/business/finance/portugal-launches-first-open-source-ai-model-joining-europes-sovereignty-push-2026-07-01/",
"day": "1. Juli 2026",
"context": "nicht angegeben",
"note": "Reuters-Datum aus Quellen-URL"
}
];
const monthNames = {
"01": "Januar", "02": "Februar", "03": "März", "04": "April", "05": "Mai", "06": "Juni",
"07": "Juli", "08": "August", "09": "September", "10": "Oktober", "11": "November", "12": "Dezember"
};
const quarterName = (month) => {
const [year, mm] = month.split("-");
return `${year} Q${Math.ceil(Number(mm) / 3)}`;
};
const monthLabel = (month) => {
const [year, mm] = month.split("-");
return `${monthNames[mm]} ${year}`;
};
const byQuarter = releases.reduce((acc, release) => {
const q = quarterName(release.month);
acc[q] ??= {};
acc[q][release.month] ??= [];
acc[q][release.month].push(release);
return acc;
}, {});
const timeline = document.querySelector("#timeline");
const empty = document.querySelector("#empty");
function render(filter = "") {
const needle = filter.trim().toLowerCase();
timeline.innerHTML = "";
let visible = 0;
Object.entries(byQuarter).forEach(([quarter, months], qIndex) => {
const quarterEntries = Object.values(months).flat();
const filteredQuarterEntries = quarterEntries.filter(matches(needle));
if (!filteredQuarterEntries.length) return;
const qDetails = document.createElement("details");
qDetails.className = "quarter";
qDetails.open = qIndex >= Object.keys(byQuarter).length - 2 || Boolean(needle);
qDetails.innerHTML = `
<summary>
<span class="quarter-title"><strong>${quarter}</strong><span>${filteredQuarterEntries.length} Releases</span></span>
<span class="quarter-meta">${Object.keys(months).map(monthLabel).join(" · ")}</span>
</summary>
`;
Object.entries(months).forEach(([month, entries]) => {
const filtered = entries.filter(matches(needle));
if (!filtered.length) return;
visible += filtered.length;
const mDetails = document.createElement("details");
mDetails.className = "month";
mDetails.open = qDetails.open || Boolean(needle);
mDetails.innerHTML = `
<summary>
<span class="month-name">${monthLabel(month)}</span>
<span class="quarter-meta">${filtered.length} Einträge</span>
</summary>
<div class="rows">
${filtered.map(releaseTemplate).join("")}
</div>
`;
qDetails.appendChild(mDetails);
});
timeline.appendChild(qDetails);
});
empty.style.display = visible ? "none" : "block";
document.querySelector("#stat-models").textContent = releases.length;
document.querySelector("#stat-months").textContent = new Set(releases.map((r) => r.month)).size;
document.querySelector("#stat-context").textContent = releases.filter((r) => r.context && !/^nicht/.test(r.context)).length;
}
function matches(needle) {
return (release) => !needle || [release.month, release.model, release.org, release.type, release.desc, release.source, release.context, release.day]
.filter(Boolean)
.join(" ")
.toLowerCase()
.includes(needle);
}
function releaseTemplate(release) {
const dateDisplay = release.day && release.day !== "Tag nicht veröffentlicht" ? release.day : monthLabel(release.month);
const context = release.context ? `<span class="pill context">Context: ${escapeHtml(release.context)}</span>` : "";
const note = release.note ? `<span class="pill">${escapeHtml(release.note)}</span>` : "";
return `
<article class="release">
<div class="date-cell">
<span class="dot" aria-hidden="true"></span>
<span class="date-main">${escapeHtml(dateDisplay)}</span>
</div>
<div class="model">
<h3>${escapeHtml(release.model)}</h3>
<div class="kind">${escapeHtml(release.type)}</div>
</div>
<div class="org">${escapeHtml(release.org)}</div>
<div class="desc-wrap">
<p class="desc">${escapeHtml(release.desc)}</p>
<div class="meta">
${context}
${note}
<a class="pill source" href="${release.url}" target="_blank" rel="noreferrer">Quelle: ${escapeHtml(release.source)}</a>
</div>
</div>
</article>
`;
}
function escapeHtml(value) {
return String(value).replace(/[&<>"']/g, (char) => ({
"&": "&",
"<": "<",
">": ">",
'"': """,
"'": "'"
}[char]));
}
document.querySelector("#search").addEventListener("input", (event) => render(event.target.value));
document.querySelector("#expandAll").addEventListener("click", () => document.querySelectorAll("details").forEach((detail) => { detail.open = true; }));
document.querySelector("#collapseAll").addEventListener("click", () => document.querySelectorAll("details").forEach((detail) => { detail.open = false; }));
render();
</script>
</body>
</html>