Claude Opus 4.6

Claude Opus 4.6, released on 5 February 2026, was Anthropic's flagship for the first half of that year and the bridge between the Claude 4 generation and the Claude 5 family that followed in the summer. Its headline change was reach: this was the first Opus-class model with a one-million-token context window, offered in beta, alongside output of up to 128,000 tokens. Long context is only useful if the model can actually find things in it, and Anthropic published the numbers to back it — 93 percent on the eight-needle MRCR v2 retrieval benchmark at 256,000 tokens, falling to 76 percent at the full million. Pricing stayed at the level Opus 4.5 established, five and twenty-five dollars per million tokens, with a higher tier of ten and thirty-seven fifty for requests whose input exceeds 200,000 tokens. The second change was how the model reasons. Opus 4.6 moved from an on/off extended-thinking switch to adaptive thinking with four effort levels, the scheme Claude Opus 5 later inherited and extended to five. Anthropic reported a lead of roughly 190 Elo points over Opus 4.5 on GDPval-AA, top place on the Terminal-Bench 2.0 agentic coding benchmark, and 90.2 percent on BigLaw Bench for legal reasoning. It shipped with agent teams in Claude Code and remains available as a previous-generation model; wujec.ai keeps this profile next to Opus 4.5 and Opus 5 rather than folding them together.

▸Videos

No videos yet.

▸Reviews