Anthropic ships Claude Sonnet 5.5: half of Opus 5.5's price, matches it on most benchmarks

AGI Hunt · wechat · 2026-09-29

One week after Opus 5.5, Anthropic released Claude Sonnet 5.5 — priced the same as Sonnet 5 ($2/$10 per M tokens), 30%+ faster, with up to 30% lower token cost per task, positioned as a cheaper companion to Opus 5.5.

Benchmarks: Terminal-Bench 4.0 jumped from 10.3% to 70.6%, beating Opus 5.5's 66.4%; Chartography quadrupled to 61.6%; GDPval-AA within 2 points of Opus 5.5, while GPT-6 Sol trails on most tasks.

Effort tiers are the highlight: Low/Medium settings beat Sonnet 5's max at 1/10th the cost (AA-Briefcase: $1.6/task at Elo 1460 vs $14 at 1360). A caveat — FrontierCode scores dropped from 52.1% at Xhigh to 46.2% at Max, as the model over-delegates to code-review subagents, causing timeouts or out-of-scope edits.

Customer numbers: Balyasny cut tokens from 497K to 121K on finance tasks; Box saw 2.4x speed; Atlassian up to 30% faster. First Sonnet with cyber safeguards, anti-distillation classifiers, and preserved thinking; also the first Sonnet to beat Pokémon Red from screenshots alone. Live in Claude apps, Code, and major cloud APIs; Haiku 5.5 coming in weeks.

Original post →

Anthropic ships Claude Sonnet 5.5: half of Opus 5.5's price, matches it on most benchmarks

More from Models