AI developer tools · What shipped

Claude Fable 5 achieved a new high score of 161 on the Epoch Capabilities Index, surpassing GPT-5.5 Pro by 1 point and marking the first time Anthropic has led the index in over a year. Epoch AI recently expanded its benchmarking hub by tracking 13 new evaluations, with 7 incorporated into the Capabilities Index, and added nine external benchmarks spanning agentic work, cybersecurity, algorithm engineering, forecasting, and research-level physics.

Qwen3-Coder and Qwen3-Coder-Next have been released as open-weight coding models for agentic software work. Qwen3-Coder-Next is an 80B MoE model that activates 3B parameters per forward pass, designed for local coding agents and lower-cost repo-level workflows, with a research paper available. Kimi K2.7 Code is a 1-trillion parameter open-weight MoE model by Moonshot AI featuring 256K context window and autonomous multi-file execution capabilities. Devstral 2, released by Mistral AI, is an open-weight coding model family with dense (123B) and local-friendly (24B) variants, both with 256K context windows, purpose-built for multi-file codebase manipulation and tool use in autonomous software engineering agents.

Grok 4.5 ships with 80 TPS speeds, a 500k context window, and benchmark results positioning it at Opus and GPT 5.5 tier, priced at $2/$6 per million tokens. Claude Sonnet 5 lands near Opus 4.8 on some tasks at a lower price point but runs approximately 30 percent more tokens due to a new tokenizer. GLM 5.2, an open-weight model, is positioned as a rival to GPT-5.5 with published benchmarks and pricing details.