New AI models and model news that matter if you build software. Every item links to the lab's own announcement, docs or model card, and shows its publication date and the date we checked it. We don't list unverified claims from social media.
Google DeepMindFrontier model
Gemini 4 Argon
Google's new frontier model, generally available via the Gemini API (paid tier) in AI Studio and Android Studio, and in Google Antigravity for AI Ultra subscribers. 1M unified input-output token limit with a new continuation protocol. Introductory API price $2 / $10 per 1M input/output tokens until Nov 30, 2026, then $4 / $20.
For builders: Relevant if you run long agentic coding jobs: check the continuation-protocol docs before swapping it into an existing agent loop.
Meta's most capable models, aimed at personal and professional agentic tasks. Muse Wonder is available through the Meta Model API and Muse Code; Muse Wonder Full is available to Muse Code subscribers.
For builders: Available to developers via the Meta Model API. Meta says a fuller safety report is coming later.
GPT-6 with 'Intelligent UI' (interactive responses) rolls out in the ChatGPT Chat tab: GPT-6 Sol for paid tiers, GPT-6 Luna for Free and Go. OpenAI says the models powering Work and Codex are not changing in this release.
For builders: Mostly a ChatGPT change. Per OpenAI, Codex models are unchanged by this release.
Public preview API on Mistral Studio. 1T-parameter multimodal mixture-of-experts model with 52B active parameters. Mistral says the open weights will be released by the end of October 2026.
For builders: The weights are not out yet (promised by end of month). Until then it's API preview only.
Open (Apache 2.0) 740M-parameter multimodal embedding model for text, code, images, audio and video. 8K context. Weights on Hugging Face and Kaggle. Google reports MTEB Code improving from 68.76 to 78.68.
For builders: Useful for local codebase indexing and retrieval for coding agents. Runs on-device; supported by llama.cpp, Ollama, vLLM and others, per Google.