Daily AI news, sourcedUpdates / Models
Updates Models · 505 items · newest issue first · page 5 of 26
AreaAllModelsAgents & codingApps & productsChips & infraResearchPricing & limitsBusiness & fundingPolicy & safetyOther
- QUASAR releases 4-bit QAT checkpoints for Qwen3.5-4B, Gemma 4 and Muse Glimmer 30B that its authors say stay closer to BF16 than Google's and NVIDIA's 4-bit buildsOfficialQUASAR (Cornell, Together AI)Oct 16Models
- Meta releases Muse Spark Open, an 816B open-weights multimodal model with 42B active parameters and 1M-token context that runs on two M5 Ultra Mac Studios
- OpenRouter launches Private Endpoints so enterprises can route fine-tuned models and reserved capacity through its APIOfficialOpenRouterOct 16Models
- Self-hosted Intelligence publishes a round-trip translation leaderboard for 15 self-hosted models across quantizations, filterable by context and memoryUnconfirmedr/LocalLLaMAOct 16Models
- NVIDIA open-weights Nemotron 3.5 Super VL, a 120B MoE multimodal agent model with 12B active parameters and day-0 vLLM supportOfficialNVIDIAOct 16Models
- Mistral releases Timestral, open time-series foundation models from 5M to 209M that lead GIFT-Eval, TIME and fev-benchOfficialMistral AIOct 16Models
- Black Forest Labs open-weights FLUX 3 Image Klein, a 5.6B text-to-image and editing model with up to 10 references and bounding-box control
- Epoch AI puts Gemini 4 Argon at 165 on its Capabilities Index, fifth of 260 models behind Claude Opus 5.5 at 167OfficialEpoch AIOct 15Models
- Researcher benchmarks exact capture of dictated phone numbers in German and French; Whisper scores 34% exactUnconfirmedr/MachineLearningOct 15Models
- IFM releases Jais 3, six open Arabic-centric agent models from 0.9B to 375B built on K2 HorizonOfficialIFMOct 15Models
- LocalLLaMA user posts KLD and perplexity for 21 Qwen3.8 27B KV-quant setups that fit 16GB VRAMUnconfirmedr/LocalLLaMAOct 15Models
- NineNineSix releases TYGER, an open-weights ASR model with full-duplex turn control on NemotronOfficialNineNineSixOct 15Models
- Microsoft details MAI Code 1.1 Flash, a 137B MoE local coding model for GitHub Copilot on WindowsOfficialMicrosoftOct 15Models
- HDTenEightyP releases GPT-USENET-6, a 0.4B model trained on USENET posts, web scrapes and fictionOfficialHugging FaceOct 15Models
- TypeSafe AI releases Jev 1.14 with experimental image input, a Box type and opt-in 262k context
- Kiro retires Claude Sonnet 4 as of October 14 and points users to Claude Sonnet 5.5
- Specific Labs releases Real-SWE 1.1, a 25-task coding-agent benchmark on licensed private enterprise codebasesOfficialSpecific LabsOct 15Models
- OpenSpline open-sources a self-hosted stack that turns a portrait and agent audio into a realtime lip-synced avatarOfficialOpenSplineOct 15Models
- StepFun publishes Step-5-Preview-NVFP4, an 8-bit ModelOpt NVFP4 build of its 600B MoE agent model on Hugging Face
- Mirelo releases SFX 2.0, generating sound effects for videos up to 10 minutes longOfficialMireloOct 15Models