AI model release tracker
Recent AI model releases developers can build on, with the announcement date, context window and whether the weights are open, each linked to the Tech AI Wire story that covered it.
Updated · All trackers
| Model | Vendor | Announced | Context window | Open weights | Reported in |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Anthropic | 2026-09-01 | 1M tokens (128K max output) | No | Read the story |
| Claude Mythos 5.1 | Anthropic | 2026-09-01 | not reported | No (invitation-only) | Read the story |
| Gemini 3.8 Flash | 2026-09-02 | 1M tokens | No | Read the story | |
| Muse Spark 1.3 | Meta | September 2026 | 1M tokens | No (proprietary; 1.3 weights undecided) | Read the story |
| GLM-5.3-Flash | Z.ai | August 2026 | 1M tokens | Yes | Read the story |
| Granite 4.2 | IBM | August 2026 | not reported | Yes | Read the story |
| WeatherNext 3 | Google DeepMind | 2026-09-03 | n/a (weather model) | not stated | Read the story |
| Koa | Salesforce | 2026-09-15 | not reported | No (proprietary; Salesforce infrastructure only) | Read the story |
| Jev | TypeSafe AI | 2026-09-15 | not reported | not reported | Read the story |
| Qwen3.8-Omni-Flash | Alibaba | 2026-09-18 | 1M tokens (991K input, 131K output) | No (API only) | Read the story |
| Laya | Convai Innovations | not reported | 512 tokens (multilingual variant: 1,024) | Yes (Apache 2.0) | Read the story |
| Qwen-Image-2.1 | Alibaba | 2026-09-20 | not reported | Yes (research license; no commercial use without permission) | Read the story |
| MiMo-V2.6-Pro | Xiaomi | September 2026 | 1,048,576 tokens | Yes (MIT) | Read the story |
| MiMo-V2.6-Flash | Xiaomi | September 2026 | 1,048,576 tokens | Yes (MIT) | Read the story |
| Claude Opus 5.5 | Anthropic | 2026-09-22 | 1M tokens (128K max output) | not reported | Read the story |
| LFM2.5-VL-DSpark | Liquid AI | 2026-09-24 | not reported | not reported | Read the story |
| Holo4-27B | H Company | 2026-09-28 | 262,144 tokens | Yes (CC BY-NC 4.0, noncommercial) | Read the story |
| Holo4-35B-A3B | H Company | 2026-09-28 | 262,144 tokens | Yes (Apache 2.0) | Read the story |
| Claude Sonnet 5.5 | Anthropic | 2026-09-28 | 1M tokens (128K max output) | not reported | Read the story |
| Kumo Tabular | NVIDIA | 2026-09-29 | not reported | Yes (OpenMDW-1.1, commercial use) | Read the story |
| GPT-6.1 Sol | OpenAI | 2026-09-29 | 1,050,000 tokens (128K max output) | not reported | Read the story |
| Gemini 4 Argon | 2026-09-30 | 1,048,576 tokens (shared by input, thinking and output) | not reported | Read the story | |
| Clef | Cloudflare | 2026-10-01 | 64K tokens | Yes (Apache 2.0) | Read the story |
| Clef-flash | Cloudflare | 2026-10-01 | 64K tokens | Yes (Apache 2.0) | Read the story |
| Strands Decider 2B | AWS (Strands Labs) | 2026-10-01 | not reported | Yes (Apache 2.0) | Read the story |
| AstaBrief 8B | Ai2 | 2026-10-02 | 16,000 tokens | Yes (Apache 2.0) | Read the story |
| Kolibri 1 | Aleph Alpha | 2026-10-03 | 1,048,576 tokens (longest training stage 262,144) | Yes (Apache 2.0) | Read the story |
| Beam | Reflection AI | 2026-10-05 | 1M tokens | Yes (Apache 2.0) | Read the story |
| EmbeddingGemma 2 | Google DeepMind | 2026-10-06 | 8,192 tokens | Yes (Apache 2.0) | Read the story |
| Mistral Large 4 | Mistral AI | 2026-10-06 | 1M tokens | Not yet (promised by end of October; license not named) | Read the story |
| Nano Banana 2.1 | 2026-10-06 | not reported | not reported | Read the story | |
| Claude Haiku 5.5 | Anthropic | 2026-10-07 | not reported | not reported | Read the story |
| GLM-Legal | Z.ai | 2026-10-12 | not reported | No (gated API beta; weight release not disclosed) | Read the story |
| Muse Wonder | Meta | 2026-10-12 | 1M tokens | No (closed weights) | Read the story |
| Muse Wonder Full | Meta | 2026-10-12 | 1M tokens | No (closed weights; Muse Code only) | Read the story |
| Microsoft-Decision-1 | Microsoft | 2026-10-09 | not reported | not reported (post-trained from Qwen3.5-9B) | Read the story |
| pplx-decider | Perplexity | 2026-10-01 | not reported | Yes (Apache 2.0) | Read the story |
| d1-3B | Liquid AI | 2026-10-07 | not reported | Yes (license not stated) | Read the story |
| d1-omni-600M | Liquid AI | 2026-10-07 | not reported | Yes (license not stated) | Read the story |
| Clef-omni | Cloudflare | 2026-10-09 | not reported | Yes (license not stated) | Read the story |
| Muse Spark Open | Meta | 2026-10-15 | 1M tokens | Yes (Muse Model License; limited commercial terms) | Read the story |
| Grok Voice 3.0 | xAI | 2026-10-14 | not reported | not reported | Read the story |
| Jais 3 (0.9B to 375B) | MBZUAI IFM and Inception42 | 2026-10-15 | 512,000 tokens | Yes (Apache 2.0) | Read the story |
| Jais Speech | MBZUAI IFM and Inception42 | 2026-10-15 | not reported | not reported | Read the story |
| Nemotron 3.5 Super VL | Nvidia | 2026-10-15 | Up to 1M tokens | Yes (OpenMDW 1.1) | Read the story |
| Nemotron 3.5 Bolt 74B-A9B | Nvidia | 2026-10-15 | not reported | Yes (license not stated) | Read the story |
| Nemotron 3.5 Bolt E4B | Nvidia | 2026-10-15 | not reported | not reported | Read the story |
| FLUX 3 Image Klein | Black Forest Labs | 2026-10-15 | not reported | Yes (FLUX Kommunity License) | Read the story |
| DeepSeek-V4.1-Flash-1017 | DeepSeek | 2026-10-17 | 1M tokens | Yes (MIT) | Read the story |
| Ling-3.1-flash | Ant Group (inclusionAI) | October 2026 | 1M tokens | Yes (MIT) | Read the story |
| Glimpse | Elorian | 2026-10-16 | not reported | not reported | Read the story |
| Step 5 Preview | StepFun | 2026-09-20 | 1M tokens | Yes (StepFun Model License 1.0) | Read the story |
| Clairic | Cloudflare | 2026-10-19 | 65,536 tokens | Not yet (promised "soon", no date) | Read the story |
| M3.1 Flash | MiniMax | 2026-10-19 | 1M tokens | Not yet (promised later, no date) | Read the story |
| Voxtral 3 (speech-to-text and text-to-speech family) | Mistral AI | 2026-10-20 | not reported | Not stated by Mistral (one report calls it open-weight) | Read the story |
| Mercury 3 | Inception | 2026-10-19 | 1M tokens | not reported | Read the story |
| Eleven Realtime v1 Max and v1 Mini (speech-to-speech) | ElevenLabs | 2026-10-20 | not reported | not reported | Read the story |
Changelog
- : Added Mistral Voxtral 3, launched October 20, 2026. Mistral states no license, so the open-weights column records that one report calls it open-weight. Added Inception Mercury 3, a diffusion language model launched October 19, 2026, with a 1M-token context; its article does not report whether the weights are open. Added ElevenLabs Eleven Realtime v1 Max and v1 Mini, speech-to-speech voice models launched October 20, 2026; the article reports neither a context window nor whether the weights are open. Reviewed the PHP 8.5.12, DuckDB 2.0 and Pigment stories too; none announces a model.
- : Added Cloudflare Clairic, previewed October 19, 2026. Reviewed the AMD GAIA 0.25 and Tenstorrent Series E stories too; neither announces a model, so neither adds a row. Added MiniMax M3.1 Flash, rolled out October 19, 2026.
- : Updated Beam: Reflection released its weights under Apache 2.0 on October 16, 2026, so the open-weights column now reads Yes.
- : Added StepFun Step 5 Preview, announced September 20, 2026, whose open weights are now on Hugging Face after being promised for October 15.
- : Added Glimpse, Elorian's first model, released on October 16, 2026 as an API research preview. Elorian has not published its context window, and the story does not say whether its weights are open.
- : Added DeepSeek-V4.1-Flash-1017, the October 17, 2026 update to V4.1 Flash with MIT-licensed weights and a 1M-token context, and Ling-3.1-flash, Ant Group's 560B open-weight model under the MIT license, whose Hugging Face repository dates from October 13, 2026.
- : Nemotron 3.5 Bolt 74B-A9B now shows open weights, citing the RTX Spark launch story, which reports Nvidia's statement that the model's training data, weights and recipes are open. China's draft AI export rules name no model, so no row was added for them.
- : Added Nemotron 3.5 Super VL, Bolt 74B-A9B and Bolt E4B, the three Nemotron 3.5 models Nvidia announced on October 15, 2026, and FLUX 3 Image Klein, the 5.6B open-weight image model Black Forest Labs released the same day. Microsoft's MAI-Base-2 has finished pre-training but is not released, so no row was added for it.
- : Added Jais 3, the six open Arabic models from 0.9B to 375B parameters that MBZUAI IFM and Inception42 released on October 15, 2026, and its Jais Speech recognition model.
- : Added Grok Voice 3.0, the speech-to-speech model xAI released in its Voice Agent API on October 14, 2026. The Amazon Bee, CERN-Mistral and Sony Walkman stories announced no AI model release, so no rows were added for them.
- : Added Muse Spark Open, the open-weight Muse Spark model Meta released on October 15, 2026 (816B parameters, 42B active, 1M-token context). The Cohere-AISLE deal announced no model, so no row was added for it.
- : Added four decision models from the AI decision models explainer: Perplexity pplx-decider (October 1), Liquid AI d1-3B and d1-omni-600M (October 7) and Cloudflare Clef-omni (October 9), 2026.
- : Added Microsoft-Decision-1, the decision-scoring model Microsoft released in Foundry on October 9, 2026. The Anthropic distillation-defenses story announced no new model, so no row was added for it.
- : Gemini 4 Argon became generally available on October 13, 2026; its row now carries the 1,048,576-token context window that article reports.
- : Added Muse Wonder and Muse Wonder Full, the frontier models Meta released on October 12, 2026. The TypeSafe AI funding story reported no new model, so Jev's existing row is unchanged.
- : Added GLM-Legal, Z.ai's legal-domain model, announced October 12, 2026. Its own context window is not reported; it is offered through a gated API beta.
- : Added 25 models from the launch stories published September 15 to October 9, 2026. Cells repeat only what the linked article states; anything it does not state, including open weights for closed models, is marked not reported.
- : Tracker created from the model-launch stories published between August 27 and September 4, 2026. Every cell repeats what the linked article reported; unknowns are marked, never guessed.
Sources
- Claude Fable 5.1 cuts cache reads 75% and keeps token prices flat - Tech AI Wire
- Gemini 3.8 Flash and Muse Spark 1.3 launch week - Tech AI Wire
- Z.ai and IBM ship open-weight reasoning models a day apart - Tech AI Wire
- Google's WeatherNext 3 brings hourly, 5km weather forecasts - Tech AI Wire
- Salesforce Koa builds on open weights but ships closed - Tech AI Wire
- TypeSafe's Jev returns typed decisions, not text - Tech AI Wire
- Qwen3.8-Omni-Flash cuts audio input costs by 98% - Tech AI Wire
- Laya is a 421M open-weights answer to Jev - Tech AI Wire
- Qwen-Image-2.1 ships 7B weights under a research license - Tech AI Wire
- Xiaomi MiMo-V2.6-Pro ships 1T open weights under MIT - Tech AI Wire
- Claude Opus 5.5 cuts output price to $20 per million - Tech AI Wire
- Liquid AI LFM2.5-VL-DSpark speeds up its 3B vision model - Tech AI Wire
- Holo4 open-weight agent models score 61.7% on OSWorld 2.0 - Tech AI Wire
- Claude Sonnet 5.5 nears Opus 5.5 at half the price - Tech AI Wire
- NVIDIA Kumo Tabular tops TabArena with open weights - Tech AI Wire
- GPT-6.1 Sol nearly matches Astra at one-fifth the cost - Tech AI Wire
- Gemini 4 Argon opens to cyber defenders before developers - Tech AI Wire
- Cloudflare Clef: open-weight models that return decisions - Tech AI Wire
- Strands Decider 2B makes agent choices in 115 ms - Tech AI Wire
- AstaBrief 8B: Ai2's open model writes cited reports - Tech AI Wire
- Aleph Alpha Kolibri 1: open German-English MoE model - Tech AI Wire
- Reflection AI launches Beam, a 501B open-weight model - Tech AI Wire
- EmbeddingGemma 2 puts multimodal search on a phone - Tech AI Wire
- Mistral Large 4 launches with 1.05T parameters - Tech AI Wire
- Nano Banana 2.1 halves the price of Gemini images - Tech AI Wire
- Claude Haiku 5.5 cuts prices 90% below Haiku 4.5 - Tech AI Wire
- Z.ai launches GLM-Legal, an AI model for legal work - Tech AI Wire
- Meta Muse Wonder launches as its first frontier model - Tech AI Wire
- Gemini 4 Argon is now generally available at $2/$10 - Tech AI Wire
- Microsoft-Decision-1 model scores choices, not text - Tech AI Wire
- AI decision models, explained - Tech AI Wire
- Meta Muse Spark Open ships 816B open weights - Tech AI Wire
- xAI Grok Voice 3.0 ships at $0.08 a minute - Tech AI Wire
- Jais 3 brings six open Arabic models, up to 375B - Tech AI Wire
- Nvidia Nemotron 3.5 Super VL ships open 120B weights - Tech AI Wire
- FLUX 3 Image Klein ships as a 5.6B open-weight model - Tech AI Wire
- Nvidia RTX Spark laptops go on sale from $2,599 - Tech AI Wire
- DeepSeek V4.1-Flash-1017 raises agent scores, keeps price - Tech AI Wire
- Ling-3.1-flash: Ant Group opens 560B model under MIT - Tech AI Wire
- Elorian Glimpse model rechecks images while it reasons - Tech AI Wire
- StepFun Step 5 Preview: 600B open-weight agent model - Tech AI Wire
- Reflection Beam open weights land under Apache 2.0 - Tech AI Wire
- Cloudflare Clairic model returns typed values, not text - Tech AI Wire
- MiniMax M3.1 Flash rolls out, claims 70% cheaper agents - Tech AI Wire
- Mistral Voxtral 3 adds voice agents from $0.0035 a minute - Tech AI Wire
- Inception Mercury 3 diffusion LLM claims 1,068 tokens/sec - Tech AI Wire
- ElevenLabs Eleven Realtime voice agents reply in 570 ms - Tech AI Wire
The weekly digest
One email every Friday with the week's top stories from all six desks: AI, dev tools, coding, the tech industry, startups and what's next. Free, no spam.