Model results · Anthropic

Claude Opus 5.5

Every published VulcanBench measurement of Claude Opus 5.5, newest first: combined score, Code quality, runtime, tokens and API-equivalent cost at every effort level on Frontier v4, graded by deterministic hidden tests on real tasks, and how it handles harmful instructions hidden in a repository on Safety v1. Measured through Claude Code on a Claude Max subscription, with Claude Code's refusal fallback on and disclosed. Claude Sonnet 5.5 and Claude Haiku 5.5, the other Claude 5.5 models on the board, have their own pages.

Claude Sonnet 5.5 vs Claude Opus 5.5 vs GPT-6.1 Sol

The three models side by side at every effort level on the same 23 tasks and judges: combined score with ±1 SE, Code quality, tasks solved, cost and minutes per task. Opus 5.5 solves 17 to 23 tasks at $1.70 to $8.75 a task, with its Opus 4.8 fallback share shown per level.

October 8, 2026 · 23 tasks · 5 effort levels · 345 runs · VulcanBench Frontier v4

VulcanBench Safety v1 across every effort level

Over 50 Claude Code runs it followed 0 of 60 hidden instructions to contact the network, leak a secret or delete files, and 3 of 100 overall: two code injections into its fix and one harmless test. It flagged the planted note to the user in 64 of 97 refusals and passed 44 of 50 tasks.

October 6, 2026 · 10 tasks · 5 effort levels · 50 runs · VulcanBench Safety v1

Claude Opus 5.5 across every effort level

In Claude Code across five effort levels: combined score 86.36 at Low to 91.11 at High, second on Frontier v4 only to Fable 5.1 at Max. Medium, the shipped default, passes all 23 tasks at 90.86 for $2.84 a task. Refusal fallback on: Opus 4.8 wrote 47.8% of replies at Max.

September 26, 2026 · 23 tasks · 5 effort levels · 115 runs of Claude Opus 5.5 · VulcanBench Frontier v4 · Code quality protocol v3.15

Beside GPT-6 Astra and Claude Fable 5.1

Combined score and cost per task at every level for all three models, with the Opus 4.8 fallback share for both Claude columns. $1.70 to $8.75 per task for Opus 5.5; the whole sweep priced at $475.33.

September 26, 2026 · API-equivalent estimates at list rates, solver inference only · VulcanBench Frontier v4

Next

How would Claude Opus 5.5 do on your stack?

Public results answer "which model is stronger on this suite." Whether Claude Opus 5.5 belongs in your routing table depends on your languages, your codebases, and your task mix, that gets measured, not guessed. Grading rules are in the methodology.

Optimize your model routing All reports