Claude Fable 5.2 benchmarks vs GPT-6 Astra, Opus 5.5 and Gemini 4 Carbon

Updated 2026-11-09. Anthropic-reported scores, clearly labeled.

Quick answer

Anthropic's launch table puts Claude Fable 5.2 on top in 10 of 12 benchmarks, including 49.1% on FrontierCode 2.0 (GPT-6 Astra 42.8%) and 69.9% on FrontierSWE v2 (Astra 65.5%). Gemini 4 Carbon wins AutomationBench. Every score is Anthropic-reported, not independently tested.

All scores below are reported by Anthropic in its Nov 9, 2026 launch post and system card summary. LimitDigest did not run them. Fable 5.2, Fable 5.1 and Opus 5.5 are scored with safeguards on, at each model's highest available effort unless noted. Bold marks the top score in each row. Gemini 4 Carbon has no public price yet.

Benchmark	Fable 5.2	Fable 5.1	Opus 5.5	GPT-6 Astra	Gemini 4 Carbon
FrontierCode 2.0 Agentic coding	49.1%	32.4%	45.5%	42.8%	41.1%
FrontierSWE v2 Agentic coding	69.9%	56.3%	62.3%	65.5%	61.2%
AutomationBench Business workflows	51.4%	31.4%	42.5%	41.4%	53.9%
Terminal-Bench-Science 0.1 Scientific research	74.6%	52.6%	58.7%	68.1%	64.8%
WANDR (relaxed) Agentic search	82.1%	68.7%	72.3%	not listed	not listed
OSWorld 2.1 (offline, partial) Computer use	89.3%	83.5%	87.2%	79.5%	81.5%
RoboDojo Sim Robotics	64.9%	27.6%	39.4%	29.8%	not listed
AA-Briefcase v1.1 (rubric) Knowledge work	66.1%	61.5%	61.2%	52.0%	61.8%
AA-Briefcase v1.1 (Elo) Knowledge work	1785	1669	1800	1569	1611
Remote Labor Index Knowledge work	41.3%	17.9%	21.3%	20.8%	not listed
Humanity's Last Exam (tools) Reasoning	74.2%	65.6%	67.7%	57.2%	not listed
Chartography (no tools) Vision	73.4%	44.8%	64.4%	71.0%	72.5%

Where Fable 5.2 leads

Fable 5.2 has the top score in 10 of the 12 rows in Anthropic's table. The biggest jumps over Fable 5.1 are in robotics (64.9% vs 27.6% on RoboDojo Sim), remote labor (41.3% vs 17.9%) and chart reading (73.4% vs 44.8%). Against OpenAI's flagship, it beats GPT-6 Astra on both coding tests: 49.1% vs 42.8% on FrontierCode 2.0 and 69.9% vs 65.5% on FrontierSWE v2.

Where it does not

Two rows go elsewhere. Gemini 4 Carbon scores 53.9% on AutomationBench, ahead of Fable 5.2 at 51.4%. On AA-Briefcase, Opus 5.5 keeps a higher Elo (1800 vs 1785) even though Fable 5.2 wins the rubric score. Anthropic also says that in AI research areas such as accelerator kernel work, Fable 5.2 performs at about Opus 5.5's level by design.

Cost per task, not just score

Anthropic's charts plot accuracy against cost. Its claims: on FrontierCode 2.0, Fable 5.2 at High effort scores 45.3%, level with Opus 5.5's best (45.5%) for roughly half the cost per task. On WANDR, Fable 5.2 at Medium matches Opus 5.5's top score for about 40% of the cost. On Humanity's Last Exam, Low effort matches Opus 5.5's top score for under a tenth of the cost. On BenchCAD 2.1 hardware design, Fable 5.2 at Max effort scores 83.5, against 65.3 for Opus 5.5 and 44.4 for Fable 5.1.

Mythos 5.2: the security and biology numbers

Benchmark (Mythos models)	Mythos 5.2	Mythos 5.1	Opus 5.5 (no safeguards)
CyScenarioBench	76.3%	61.7%	67.6%
CVE Rediscovery	89.6%	78.3%	81.3%
De novo binder design (24 hr)	85.1%	79.1%	82.6%

These rows only exist for Mythos because Fable's safeguards block most of these tasks. Anthropic says Mythos 5.2 found 10.2% more vulnerabilities than Opus 5.5 and 14.4% more than Mythos 5.1 on CVE Rediscovery, across 1,829 CVEs in an October 2026 run.

Long-run demos

Anthropic also reports two showcase runs. A team of about 1,000 Fable 5.2 agents formalized Fermat's Last Theorem in Lean in 30 hours, while Opus 5.5 in the same setup finished 26% in 18 hours. And Fable 5.2 designed three working electronic devices, with circuit boards ready for manufacturing in under a day. These are single demonstrations, not benchmarks.

Safety numbers

In a sandbox-escape test of over 1,800 scenarios, Mythos 5.2 used credentials or files placed outside its sandbox in 3.1% of scenarios, against 3.3% for Opus 5.5, 5.1% for Sonnet 5.5, and 21% and 39% for Mythos 5.1 and Mythos 5. Anthropic says Fable 5.2 produced about 55% fewer unnecessary blocks than Fable 5.1 in early testing.

Verdict

On Anthropic's own numbers, Fable 5.2 is the strongest released model for agentic coding and long technical work, and it beats GPT-6 Astra on price too. Gemini 4 Carbon wins business automation on AutomationBench. Treat every figure here as vendor-reported until independent boards such as LMArena rank Fable 5.2.

Full launch coverage: Claude Fable 5.2 and Mythos 5.2: prices, limits and what changed.

Sources

FAQ

What are Claude Fable 5.2's benchmark scores?

Per Anthropic: 49.1% on FrontierCode 2.0, 69.9% on FrontierSWE v2, 89.3% on OSWorld 2.1, 74.2% on Humanity's Last Exam with tools and 64.9% on RoboDojo Sim.

Is Claude Fable 5.2 better than GPT-6 Astra?

On Anthropic's table, yes on every shared benchmark: for example 49.1% vs 42.8% on FrontierCode 2.0 and 89.3% vs 79.5% on OSWorld 2.1. These are Anthropic's numbers, not independent tests.

Is Fable 5.2 better than Opus 5.5?

Anthropic says Fable 5.2 is a step above Opus 5.5 in almost all capabilities it evaluated. Opus 5.5 keeps a higher AA-Briefcase Elo (1800 vs 1785) and costs half as much.

What are Claude Mythos 5.2's benchmarks?

Anthropic reports 76.3% on CyScenarioBench, 89.6% on CVE Rediscovery and 85.1% on de novo binder design, all above Mythos 5.1.

Where does Fable 5.2 lose?

On AutomationBench, where Gemini 4 Carbon scores 53.9% against Fable 5.2's 51.4%, per Anthropic's table.

More guides

Not sure a plan beats the API for you? Try the plan vs API cost calculator.