Model releases, pricing changes, compliance updates, and benchmark results — curated weekly. Everything that matters for developers, teams, and AI buyers.
Google cut Gemini 2.5 Flash pricing by 40%, making it the cheapest frontier-quality LLM API. At $0.15/$0.60 per 1M tokens, it now undercuts GPT-4o mini while delivering significantly higher quality. This is the most significant pricing shift since DeepSeek's launch.
Claude Sonnet 4.6 is Anthropic's new mid-tier model replacing Sonnet 3.7. It scores 9.2 on performance in our testing, outperforming GPT-4o (9.0) at a lower API price ($3/$15 vs $2.50/$10 per 1M tokens). Extended thinking mode is now included at no extra cost.
OpenAI's GPT-4.1 introduces a 1M token context window — previously only Gemini offered this at scale. GPT-4.1 Mini at $0.40/$1.60 per 1M tokens sits between GPT-4o mini and GPT-4o in the pricing stack. Coding benchmarks show GPT-4.1 outperforming its predecessor.
The EU AI Act began enforcement for high-risk AI systems. General-purpose AI models with >10^25 FLOPs training compute must now register with the EU AI Office. OpenAI, Google, Meta, and Anthropic have all filed compliance documentation. DeepSeek's status remains unclear for EU deployments.
Microsoft expanded GitHub Copilot access with a permanent free tier: 2,000 code completions and 50 chat messages per month. The free tier includes access to Claude Sonnet and GPT-4o (with rate limits). This puts significant pressure on Cursor and Windsurf who charge from $0/mo.
Independent testing confirms DeepSeek R1 achieves 97.3% on MATH-500 vs OpenAI o3's 97.9% — essentially matched performance at $0.55/$2.19 vs $10/$40 per 1M tokens. This is the starkest price-performance gap in reasoning model history.
OpenAI expanded Sora access from Pro-only ($200/mo) to Plus ($20/mo), with a cap of 50 video generations per month for Plus users. Video quality at 720p, up to 20 seconds. Pro users retain unlimited access and 1080p. This dramatically changes the competitive landscape for Runway and Pika.
ElevenLabs raised Series C at $3B valuation and announced support for 32 languages with real-time voice cloning. New Flash v2.5 model reduces latency to under 75ms — enabling true real-time voice applications. Enterprise plan now includes on-premises deployment option.
Get the weekly digest in your inbox
Every Thursday: pricing changes, new model releases, and benchmark results. No spam. Unsubscribe any time.
Subscribe free →