Gemini 4 Carbon takes the top spot on the Artificial Analysis Intelligence Index v5
Artificial Analysis ranks Google's Gemini 4 Carbon first on its Intelligence Index v5 with a score of 59 (high reasoning), leading four of the index's nine evaluations. It is still behind Claude Opus 5.5 on coding: 59 vs 62 on the Coding Agent Index, and 51% vs 56% on FrontierCode. Cost per index task is $7.76, about 23% below Opus 5.5 (max), and it uses fewer output tokens (123k vs 173k). The catch is reliability: its hallucination rate on AA-Omniscience rose from 4% (Gemini 4 Argon) to 11%. Carbon is still rolling out to selected users, not generally available; standard pricing is $4/$20 per million tokens, and the 50% launch discount has no confirmed end date.
Sources & depth
Mistral releases Large 4 ("Le Chonk") as open weights, with day-0 support across the main toolchains
Mistral has delivered on its promise: the weights of Mistral Large 4, a roughly 1-trillion-parameter multimodal mixture-of-experts model (about 50B active; the sources give both 49B and 52B), are now on Hugging Face. The license is a modified MIT that allows commercial use but requires a separate licence from anyone offering the model as a service. Mistral's own benchmarks put it first among open-weight models on FrontierCode 2.0 (19.7%) and roughly level with GLM-5.3 on the Artificial Analysis Coding Agent Index (48%), with Terminal-Bench 4.0 up from 28% in the preview to 41%; these are vendor-reported figures. API pricing is now $0.70 in / $2.10 out per million tokens, down from the preview's $1.36/$4.18. Running it is a serious-hardware job: 2-bit GGUF quants fit a 512 GB Apple Silicon Ultra, while the NVFP4 checkpoint needs Blackwell GPUs. Transformers v5.22.0, vLLM, llama.cpp and TRL (v1.17.0, with a single-node 8×B300 LoRA fine-tuning recipe) all support it from day one.
Sources & depth
Black Forest Labs offers FLUX 3 weights for licensing and self-hosting
Black Forest Labs now lists FLUX 3 Max, FLUX 3 Image and FLUX 3 Video (video with audio) as commercial weights that companies can license, customize, train on their own data and run on their own infrastructure, under a fixed-fee, uncapped-generation arrangement. Two smaller models are open: FLUX 3 Action, a 7B world action model, and FLUX 3 Klein Image on Hugging Face. No prices or launch date are published on the page.
Sources & depth