Daily AI news, sourcedUpdates / Models
Updates Models · 588 items · newest issue first · page 28 of 30
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- Community model claims better calibration than Jev with multi-modal abilityUnconfirmedreddit-machinelearningSep 30Models
- 500M VLM answers typed questions about images with calibrated confidenceUnconfirmedreddit-machinelearningSep 30Models
- Community tests 5 Qwen3.6-35B-A3B finetunes against the base modelUnconfirmedreddit-localllamaSep 30Models
- Paper's own Table 2 says MoGe-2 pseudo-label training mostly fails to beat MoGe-2Unconfirmedreddit-machinelearningSep 30Models
- H Company releases Holo4, open-weight computer-use models
- BAAI releases AREX-2, a 27B agent model built on Qwen3.8
- Cline shares Terminal-Bench comparison of Sonnet 5.5 and Opus 5.5Officialx-sdrznSep 30Models
- AutoTrust AI releases JEV-27B, an open-weights decision model for self-hosted AI agents
- Kuaishou's Kling 4.0 video model is due in October, with Flash in early access, per IT Home
- Anthropic releases Claude Sonnet 5.5, 30%+ faster than Sonnet 5 at the same price
- Suochen Tech and Meimeng Space release embodied models and physics benchmark, per QbitAI
- 700B-parameter GLM runs on a laptop with SSD as memory, per QbitAI
- Alibaba Qwen releases Qwen-Audio-3.1-Realtime full-duplex voice model, per MarkTechPost
- MarkTechPost verifies 20 agentic use cases of TypeSafe AI's Jev, per MarkTechPost
- Sarvam AI releases Saaras V4 speech-to-text for all 22 Indian languages, per MarkTechPost
- Supersonic Labs releases Julia 1, a 144.3M-parameter decision model that runs on CPU, per MarkTechPost
- 'Bad Apple' benchmark pushes models to their limits
- Stanford Rejuvenation A.I. Benchmark leak discussed online
- Community reports say first Qwen 4 samples approach Fable/Opus-level quality
- First days running qwen3.8-flash-next on 4x R9700