Daily AI news, sourcedUpdates / Models
Updates Models · 505 items · newest issue first · page 25 of 26
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- UltraFast Qwen3.8 Flash recipe claims 74 tok/s on one DGX Spark
- Qwen3.8 27B claimed at 95+ tok/s through 100K generation on a single 3090
- Qwen3-VL 8B on a laptop claimed to beat GPT-5.6 on messy tax-form documents
- Basis says GPT-6 Astra completes tax workbook 2x faster than GPT-5.6 Sol
- Fireworks introduces FireRouter with Opus, its first router model
- H Company releases benchmark trajectories for inspection and replay
- Community releases GSQ-RCO GGUFs for Qwen3.8-Flash-Next with expert-pruned Coder build
- IQuest-Q1: a 320B MoE model (15B active) that spots RL data bugs and builds games from prompts, per QbitAI
- Claude Opus 5.5 is now available
- IQuestLab/IQuest-Q1 · Hugging Face
- Well, if you were getting excited for a new frontier model from OpenAI anytime soon, bad news.
- Open-source vllm-jev cuts System One inference from 600ms to about 50ms
- Anthropic tightens Sonnet 5.5 thinking-block restrictions
- Cline adds Pixel Canary, a free anonymous model it says matches GPT-6 AstraOfficialCline releaseSep 28Models
- Huawei open-sources openPangu-2.0 training and RL code, per iFengReportediFeng Tech (in Chinese)Sep 28Models
- OpenAI shuts down gpt-3.5-turbo-instruct, babbage-002 and davinci-002 todayOfficialOpenAI deprecationsSep 28Models
- Anonymous model Space Bunny tops OpenRouter and OpenCode daily usage charts, per QbitAIReportedQbitAI (in Chinese)Sep 28Models
- A new Google Flow build reportedly references "Nano Banana 2.1"UnconfirmedTestingCatalogSep 28Models
- NaiveAI releases Naive-N0.5-Flash, a 309B open-weights MoE with 1M context under MITOfficialHugging FaceSep 28Models
- TextCLF launches TQ, calibration-free 4-bit quantization with an OpenAI-compatible APIOfficialTextCLF on XSep 28Models