Agents · Benchmark profile

Cybench

Professional-level Capture-the-Flag cybersecurity agent benchmark measuring unguided end-to-end task solve rates.

Ranking weightedCurrent / rollingActive7 models

Data verified 1 Aug 2026 · Methodology 2.3.0

Benchmark score on Cybench

Leaderboard · 7 models

Claude Mythos 5 leads at 100%, followed by Claude Opus 4.7 (96%) and Claude Opus 4.6 (93%).

Ranking · 1.0%AgentsCurrent

#1Closed

Claude Mythos 5

Anthropic

claude-mythos-5

100%

Overall — · Context 1M tokens

#2Closed

Claude Opus 4.7

Anthropic

claude-opus-4-7

96%

Overall #23 · Context 1M tokens

#3Closed

Claude Opus 4.6

Anthropic

claude-opus-4-6

93%

Overall #30 · Context 1M tokens

ViewModelsColours

AnthropicMetaxAI

Lumina

Cybench leaderboard

Accessible data for Cybench leaderboard
Model	Rank	Model ID	Provider	Score	Unit
Claude Mythos 5	#1	claude-mythos-5	Anthropic	100	percent
Claude Opus 4.7	#2	claude-opus-4-7	Anthropic	96	percent
Claude Opus 4.6	#3	claude-opus-4-6	Anthropic	93	percent
Muse Spark 1.1	#4	muse-spark-1-1	Meta	92.9	percent
Claude Opus 4.5	#5	claude-opus-4-5	Anthropic	82	percent
Muse Spark	#6	muse-spark	Meta	65.4	percent
Grok 4.1	#7	grok-4-1	xAI	39	percent

Benchmark score table (7 models)

One best compatible score per canonical product · higher is better

Cybench leaderboard ranked by score
Rank	Model	Provider	License	Evidence use	Score
#1	Claude Mythos 5 claude-mythos-5	Anthropic	closed	Score input	100%
#2	Claude Opus 4.7 claude-opus-4-7	Anthropic	closed	Score input	96%
#3	Claude Opus 4.6 claude-opus-4-6	Anthropic	closed	Score input	93%
#4	Muse Spark 1.1 muse-spark-1-1	Meta	closed	Score input	92.9%
#5	Claude Opus 4.5 claude-opus-4-5	Anthropic	closed	Score input	82%
#6	Muse Spark muse-spark	Meta	closed	Score input	65.4%
#7	Grok 4.1 grok-4-1	xAI	closed	Score input	39%

The top of this snapshot is led by Claude Mythos 5 at 100%; third place is 7.0 points behind. The top-7 spread is 61.0 points.

Configuration-preserving evidencePublished source rows (10)Exact source labels, configurations, versions and dated observations. Same-snapshot canonical/configuration projections are shown once.

Every distinct dated source-published row for Cybench, including exact model configuration labels
Exact published model / configuration	Canonical product	Score	Version / system	Evidence	Source / observed
Exact BenchLM registry variant Muse Spark 1.1; bulk export does not retain a complete upstream harness configuration. source label without registered configuration ID	Muse Spark 1.1	92.9%	2025Source-native system	Source carrier · no additional scoring weight source-checked	BenchLM public datasets — 2026-08-01 ↗Observed 2026-08-01 · checked 2026-08-01
Exact BenchLM registry variant Muse Spark 1.1; bulk export does not retain a complete upstream harness configuration. source label without registered configuration ID	Muse Spark 1.1	92.9%	2025Source-native system	Source carrier · no additional scoring weight source-checked	BenchLM public datasets — 2026-07-27 ↗Observed 2026-07-27 · checked 2026-07-27
Exact BenchLM registry variant Muse Spark 1.1; bulk export does not retain a complete upstream harness configuration. source label without registered configuration ID	Muse Spark 1.1	92.9%	2025Source-native system	Source carrier · no additional scoring weight source-checked	BenchLM public datasets — 21 July 2026 ↗Observed 2026-07-21 · checked 2026-07-21
Claude Mythos Preview; unguided; 35-problem subset (system card via Cybench leaderboard note ⁶) source label without registered configuration ID	Claude Mythos 5	100%	CurrentCybench / provider system-card evaluation as listed by Cybench maintainers	Score input official-leaderboard	Cybench official leaderboard ↗Observed 2026-07-15 · checked 2026-07-15
Claude Opus 4.7; unguided; 35-problem subset (system card via Cybench leaderboard note ⁸) source label without registered configuration ID	Claude Opus 4.7	96%	CurrentCybench / provider system-card evaluation as listed by Cybench maintainers	Score input official-leaderboard	Cybench official leaderboard ↗Observed 2026-07-15 · checked 2026-07-15
Claude Opus 4.6; unguided; 37-problem subset (system card via Cybench leaderboard note ⁵) source label without registered configuration ID	Claude Opus 4.6	93%	CurrentCybench / provider system-card evaluation as listed by Cybench maintainers	Score input official-leaderboard	Cybench official leaderboard ↗Observed 2026-07-15 · checked 2026-07-15
Claude Opus 4.5; unguided; 39-problem subset (system card via Cybench leaderboard note ³) source label without registered configuration ID	Claude Opus 4.5	82%	CurrentCybench / provider system-card evaluation as listed by Cybench maintainers	Score input official-leaderboard	Cybench official leaderboard ↗Observed 2026-07-15 · checked 2026-07-15
Muse Spark; unguided (Meta safety report via Cybench leaderboard note ⁷) source label without registered configuration ID	Muse Spark	65.4%	CurrentCybench / provider system-card evaluation as listed by Cybench maintainers	Score input official-leaderboard	Cybench official leaderboard ↗Observed 2026-07-15 · checked 2026-07-15
Grok 4.1 Thinking; unguided (xAI model card via Cybench leaderboard note ⁴) source label without registered configuration ID	Grok 4.1	39%	CurrentCybench / provider system-card evaluation as listed by Cybench maintainers	Score input official-leaderboard	Cybench official leaderboard ↗Observed 2026-07-15 · checked 2026-07-15
Muse Spark 1.1; provider evaluation configuration as published in the Meta Muse Spark 1.1 evaluation report (Cybench). muse-spark-1-1-xhigh	Muse Spark 1.1	92.9%	CurrentProvider evaluation harness	Public reference · not scored provider-reported	Meta AI Muse Spark 1.1 evaluation report ↗Observed 2026-07-09 · checked 2026-07-16

Compare top models

Claude Mythos 5 vs Claude Opus 4.7 Claude Opus 4.7 vs Claude Opus 4.6 Claude Opus 4.6 vs Muse Spark 1.1 Muse Spark 1.1 vs Claude Opus 4.5

About Cybench

Definition and scoring

Organisation: Stanford / Cybench authors
Category: Agents
Version: Not evaluated
Direction: higher is better
Ranking use: Weighted 1.0%
Contamination risk: Low

Professional-level Capture-the-Flag cybersecurity agent benchmark measuring unguided end-to-end task solve rates. Every genuine source row stays tied to its exact model label, configuration, benchmark version and evaluation system. The summary chart shows one best compatible score per canonical product; Score 2.0 admits only explicitly mapped, frozen protocols and keeps incompatible configurations separate.

Open benchmark source ↗

FAQ

Quick answers

What does Cybench measure?

Professional-level Capture-the-Flag cybersecurity agent benchmark measuring unguided end-to-end task solve rates.

Which model scores highest on Cybench?

Claude Mythos 5 by Anthropic currently leads with 100%.

How many models are evaluated on Cybench?

7 canonical model products have a qualifying score, backed by 10 distinct dated source rows.

Agents · Benchmark profile

Cybench

Professional-level Capture-the-Flag cybersecurity agent benchmark measuring unguided end-to-end task solve rates.

Ranking weightedCurrent / rollingActive7 models

Data verified 1 Aug 2026 · Methodology 2.3.0

Benchmark score on Cybench