Mistral Large 4
Mistral's Large 4, nicknamed Le Chonk, is a 1-trillion-parameter open-weights model with a 1M-token context that Mistral says is natively multimodal. Its first third-party score of 38 sat one point…
AI threads tagged models
Gemini 4
Google now has two Gemini 4 models: Argon, which shipped September 30 to trusted cyber defenders and then opened worldwide to developers and subscribers in mid-October, and Carbon, announced October…
LLM peer review: Sakana's Multi-Layered Review benchmark
Sakana's Multi-Layered Review benchmark is credited with catching 73% of core-claim errors on a 1,164-error corpus, though that is an unverified vendor number. Springer Nature and Google DeepMind are…
Anthropic Cyber Mission
Anthropic expanded its Cyber Verification Program to give verified security professionals broader access to its most capable models, then launched the Anthropic Cyber Mission and a free, opt-in OSS…
Claude Opus 5.5
Anthropic's Claude 5.5 family is now complete: Opus 5.5 launched September 22, Sonnet 5.5 September 28, and Haiku 5.5 October 7. Opus 5.5 took the top spot on an independent AI index and cost less…
Block's Dessa: a voice AI trained on human audio alone, one Nvidia rack, a MacBook to run it
Block, Jack Dorsey's payments company, has shown Dessa, a full-duplex speech-to-speech model it says was trained from scratch on human conversation alone, with no text and no pretrained models. Block…
NVIDIA's Vera Rubin DSX POD: modular AI-factory units and open frontier models
NVIDIA's Jensen Huang has announced Vera Rubin DSX POD modular factory units for AI infrastructure and open frontier models for custom enterprise workloads, framed as pre-packaged AI capacity. The…
IBM's Granite line: open-licensed speech models and time-series ensembles
IBM announced Granite Speech 5.0 2B, a 2-billion-parameter multilingual speech-recognition and translation model under Apache 2.0, with a claimed #2 rank on the Open ASR Leaderboard at 5.68% error…
OpenAI Decisions API
OpenAI has shipped a Decisions API that returns typed decisions from a fixed answer list in roughly 150ms on GPT-6 Luna. On October 22 it opened the API to all users after a limited…
Perplexity pplx-decider
Perplexity's pplx-decider line now includes open-weight multimodal models (v1.2 and Flash v1.2) available through its API, Hugging Face, OpenRouter, and Vercel, plus a Mac app that runs Flash locally…
GPT-6.1-Sol reasoning traces: the 'totally legible' claim
A claim says GPT-6.1-Sol's reasoning traces are totally legible, including that the trace invents German from first principles. The claim is framed as a meme, and no trace has been shown. Watch for…
Diffusion LLMs: Inception's Mercury 3 as the flagship claim
Inception has announced Mercury 3, calling it the most capable diffusion-based large language model. No pricing, availability, or independent benchmark results are known yet. Watch whether it…
Sarvam's Bulbul v4 Flash: real-time TTS in 11 Indian languages
Sarvam AI announced Bulbul V4 Flash on October 13, 2026, a real-time text-to-speech model covering 11 Indian languages. No price has been published and it has not been shown running in a live…