VulcanBench VulcanBench
Model results · OpenAI

GPT-5.5

Every published VulcanBench measurement of GPT-5.5, newest first: combined score, Code quality, runtime, tokens and API-equivalent cost at every effort level, graded by deterministic hidden tests on real tasks. The Frontier v4 runs went through the Codex CLI on a ChatGPT subscription.

Suitev4
GPT-5.5 vs. GPT-5.6 Luna across every effort level
In Codex across four effort levels: combined score 49.80 to 78.46 from Low to Extra-high and Code quality 61.75 to 67.67. Ahead of Luna at Low and Medium, behind by less than one standard error at High and Extra-high; its API has no Max level. 1 to 11 of 23 tasks passed, at 8.2 to 20.8 minutes per task.
September 15, 2026 · 23 tasks · 4 effort levels · 92 runs of GPT-5.5 · VulcanBench Frontier v4
→
Suitev4
Cost and tokens across effort levels
$1.79 to $4.56 per task at list API rates and 1.72M to 4.72M raw tokens per task across the four effort levels; 17 to 75 times Luna's cost at matched effort.
September 15, 2026 · API-equivalent estimates at list rates, solver inference only · VulcanBench Frontier v4
→
Report04
Claude Fable 5 vs Opus 4.8 vs GPT-5.5 benchmark
Four frontier configurations on 15 real engineering evals. All score 14/15 but each misses a different task; cost spans 4x.
2026-07-02
→
Next

How would GPT-5.5 do on your stack?

Public results answer "which model is stronger on this suite." Whether GPT-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