Skip to content
Spellwright News

Benchmarks

SRE-Bench, pass@4

SRE-Bench, pass@4

Other · higher is better · 1 model

Pass@4: the share of problems solved in at least one of 4 tries.

  1. Gemini 4 Carbon98.9%*

Every result

  • Gemini 4 Carbon98.9%Reported by the model's maker (Google), Nov 10, 2026

    Google reports a 98.9% pass@4 full-solve rate for Gemini 4 Carbon on SRE-Bench, a security reverse-engineering test of 262 binaries.

    Tasks: 262 binary instances across five reverse-engineering domains; full-solve rate.

    Limits: Provider-reported: Google publishes the result for its own model; it has not been independently reproduced here. Google says the scores are as of October 2026 for the model it calls Gemini 4 Carbon; equivalence to the build now served on every surface is not established. Results under the same benchmark name can be shown together as reported scores, but unmatched settings and missing methods rule out a controlled ranking.

    Google DeepMind

No published SRE-Bench, pass@4 score checked yet for: Claude Fable 5.2, Claude Mythos 5.2, Gemini 4 Argon, Claude Sonnet 5.5, Claude Opus 5.5, Claude Opus 5, Claude Sonnet 5, Gemini 2.5 Pro Experimental (March 2025), Gemini 2.0 Flash (December 2024 experimental), Qwen2.5-7B, Qwen2.5-7B-Instruct, GPT-4o mini (July 2024 release).