New releases, price changes, and ranking updates -- detected automatically from our daily data sync.
Nearly 24% of all tracked models changed prices in March. Overall AI API costs dropped 40-70% across major providers as competition heats up.
GPT-5.4 Pro sets a new FrontierMath record (50% Tiers 1-3). Both models tie at 57 points on the Intelligence Index, with Claude Opus 4.6 leading SWE-Bench at 80.8%.
GPT-5.4 Mini runs 2x faster than GPT-5 Mini and approaches full GPT-5.4 quality. Nano costs just $0.20/1M input tokens, targeting classification and subagent tasks.
Nemotron 3 Super is a 120B Mamba-Transformer hybrid with 12B active parameters, 1M context, and 7.5x higher throughput than comparable models. Open-weight under permissive license.
OpenAI released GPT-5.4 in three variants: Standard, Thinking, and Pro. The model supports up to 1.05M token context and reports a 33% reduction in factual errors vs GPT-5.2.
Gemini 3.1 Flash-Lite costs $0.25/1M input tokens, runs 2.5x faster time-to-first-token than Gemini 2.5 Flash while matching its quality.
Gemini 3.1 Pro scores 77.1% on ARC-AGI-2 — more than double its predecessor — and 80.6% on SWE-Bench Verified, all at the same price as Gemini 3 Pro.
Claude Sonnet 4.6 ships at the same price as Sonnet 4.5 with significantly improved coding, consistency, and instruction following. Shares the 1M context window of Opus 4.6.
Claude Opus 4.6 features a 1M token context window, extended thinking, and scores 80.8% on SWE-Bench Verified — the highest of any model.
Claude Opus 4.5 drops from $15 to $5/1M input tokens and $75 to $25/1M output tokens. Repositioned from boutique model to production-ready enterprise tool.