Google has released Gemini 4 Hydrogen, a fast general-purpose model for everyday coding and agentic workflows that replaces Gemini 3.8 Flash and beats the larger Gemini 4 Argon on benchmarks. It offers four reasoning levels, a one-million-token context window for both input and output, and is priced at $1.50 per million input tokens and $7.50 per million output tokens, available now via the Gemini API, Gemini app, Google Search and Google Antigravity.
Reasoning Level
推理等级
Configuration options controlling model thinking depth and output quality
Curated Intelligence
Google updated its Android Bench leaderboard, adding reasoning level annotations and new models. Claude Sonnet 5.5 leads with the highest pass rate, while GPT-6.1 Sol offers high completion rates at significantly lower costs. In contrast, Google's own Gemini 4 Argon performed poorly, ranking fifth and lagging behind competitors.
Meta released its new flagship model, Muse Wonder, which Artificial Analysis rates as equivalent to GPT-6.1 Sol. Developer Simon Willison updated his open-source tool llm-meta-ai to v0.2 to support the model, demonstrating output differences across reasoning levels using the classic 'pelican riding a bicycle' SVG generation test.
Mistral AI has released a preview of its flagship model, Mistral Large 4, featuring a Mixture-of-Experts architecture with 1 trillion total parameters and 49 billion active parameters. Currently accessible via API with open weights promised by month-end, the model scores 38 on Artificial Analysis benchmarks, trailing DeepSeek 4.1 Flash but showing significant improvement over previous generations.