Breaking

Anthropic Releases Claude Fable 5 Publicly, Tests Autonomy on Agent Arena

June 9, 2026 at 14:58 EDT

Foundation Models
AI Agents
Software Dev & Coding

On June 9, 2026, Anthropic publicly released Claude Fable 5, the first broadly available model in its top-tier "Mythos" class. The model is now available in "Agent Mode" on the agent evaluation platform Agent Arena, where its ability to autonomously carry out real-world tasks will be assessed through public sessions.

June 9, 2026 · Anthropic

Claude Fable 5 Opens the Claude 5 Era — and Steps Into the Agent Arena

The first "Mythos-class" general-release model — Anthropic's most capable model ever made broadly available — now runs real-world tasks live on the Agent Arena leaderboard, a safety-hardened counterpart to the institution-only Mythos 5.

80.3%

SWE-Bench Pro — state-of-the-art-class

token context window

$10/$50

est. price per 1M tokens (in / out)

Token Cost vs. Opus

Reported real-world usage: Fable 5 burns roughly twice the tokens of Opus on comparable work — capable, but costly.

1×

Opus

~2×

Fable 5

From Static Benchmarks → Live Task Completion

Real user tasks

research · reports · code · web · images

→

Agent Mode session

Agent Arena execution

→

Elo-style ranking

leaderboard, staged

What impressed

Built a "macOS-style web OS" — dock, draggable windows, file explorer, terminal — from a single prompt
Standout UI & front-end generation
Immediately usable on agent platforms (Akai, Asteroid)

The trade-offs

Guardrails kick in heavily on complex tasks
~2× Opus token use makes it costly
Knowledge, speed & tool access trimmed vs. Mythos 5
High-risk queries auto-routed to Opus 4.8

Built for long-horizon agents

Designed for agentic tasks spanning days, with guardrails aimed at reducing dual-use risk.

Tool use Long-horizon execution Multi-agent coordination claude.ai · API · Agent Arena

Continue reading

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