CLAUDE SONNET 4.6
1M context. 64K output. Extended + adaptive thinking. Vision. Native web search. $3/$15 per million tokens — drop-in upgrade for Sonnet 4.5 at the same price.
Claude Sonnet 4.6 is Anthropic's price-performance pick — the model that gets ~90% of Opus 4.7's coding quality at 1/5 the price. It ships with a 1M-token context window, a 64K output window (one of the largest in production), both extended and adaptive thinking modes, vision input, and native web search. Priced at $3 per million input tokens and $15 per million output, it's a drop-in upgrade for Sonnet 4.5 at the same price point — same cost, materially better refactor patches, longer context, and adaptive thinking that automatically dials reasoning depth to the task. The case for Sonnet 4.6: daily coding, multi-file refactors, technical writing, document analysis, and anything where you want Anthropic quality without paying Opus prices. The case against: the very hardest architecture decisions and voice-critical writing (Opus 4.7 still wins), agent tool loops at scale (GPT-5.5 is the most stable), and bulk codegen where every dollar counts (DeepSeek V4 is a fraction of the price).
Council AI exposes Claude Sonnet 4.6 on every paid plan inside multi-model councils. You don't manage API keys, rate limits, or per-call pricing — the model is bundled into your monthly plan budget, and the orchestrator picks it (or routes around it) based on the task. When the council runs, Claude Sonnet 4.6 produces its independent answer in parallel with the other selected models, then the Moderator synthesizes a single response that reconciles the disagreements.
For daily coding and writing, Sonnet 4.6 covers ~90% of what Opus does at 1/5 the price. Save Opus for the very hardest architecture decisions, hairy refactors, and voice-critical writing.
Same price ($3/$15), but longer context (1M), bigger output (64K), better refactor patches, and adaptive thinking that automatically dials reasoning depth. It's a drop-in upgrade — no reason to stay on 4.5.
A mode where Sonnet automatically chooses how much internal reasoning to spend based on prompt difficulty. Cheap questions stay fast and cheap; hard questions get extended thinking without you flipping a flag.
Yes — natively, via Anthropic's own web_search tool (not Brave). Council AI enables it when the council needs current information.
64,000 tokens — one of the largest in production. That's ~50,000 words in a single response, which makes it practical for long refactor patches and long-form drafts.
Yes — it's the price-performance pick for engineering work. Patches are clean enough that a senior engineer often merges them unchanged.
$3.00 per million input tokens and $15.00 per million output tokens — same as Sonnet 4.5. Council AI bundles it inside monthly plan budgets.