Breaking

Claude Fable 5 Tops Agent Arena, Beating GPT-5.5 and Opus 4.8

June 10, 2026 at 16:30 EDT

Foundation Models
AI Agents
Software Dev & Coding

Anthropic's new model Claude Fable 5 has taken the overall top spot on Arena.ai's newly launched agent evaluation board, the Agent Arena. As of June 10, 2026, it leads GPT-5.5 and Claude Opus 4.8 by what is described as the widest margin yet.

June 9, 2026 · Anthropic

Claude Fable 5 takes the top spot for autonomous agents

Anthropic's most powerful public model debuts as a "third era" built for long-horizon work — and a day later leads the new Agent Arena leaderboard by its widest margin yet on confirmed task success and praise-vs-complaint.

64.9

Intelligence Index — ranked 1st, ~5 pts over runner-up

300k+

Real user tasks & 2M+ tool calls behind Arena scoring

1M+

Context window, in tokens

Static benchmarks: Fable 5 vs the field

Higher is better. Each benchmark scaled independently.

SWE-Bench Pro · agentic coding

80.3%

Fable 5

69.2%

Opus 4.8

58.6%

GPT-5.5

FrontierCode Diamond

29.3%

Fable 5

13.4%

Opus 4.8

5.7%

GPT-5.5

What it unlocked

Flappy Bird clone with leaderboard, database & deploy in 20 min (~80k tokens)
Complex proofs completed error-free in 10 min
73% Pass@5 on DiscoverPhysics benchmark
Solved tasks that repeatedly failed on Opus 4.8

Where caution applies

Weaker steerability — refusals, hard to redirect
Fast generation can add security flaws & technical debt
Code-review precision lags Opus 4.8
Falls back to Opus 4.8 on cyber/bio/chem territory

Pricing & access

Live the same day across the Claude API, AWS Bedrock, Azure Foundry, Vertex AI and GitLab Duo. No extra charge on Pro/Max plans through June 22.

$10

per 1M input tokens

$50

per 1M output tokens

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.