Breaking

Anthropic Releases Claude Fable 5 to the Public, Posting 80.3% on SWE-bench Pro

June 9, 2026 at 13:44 EDT

Foundation Models
Software Dev & Coding
AI Agents

On June 9, 2026, Anthropic released Claude Fable 5, the first generally available model in its top "Mythos-class" tier. The company says it shows its highest-ever performance in software engineering, scientific research and knowledge work, and it became available at launch on platforms including the design-comparison service Design Arena.

June 9, 2026 · Anthropic

Claude Fable 5 Arrives: The First Public "Mythos-Class" Model

Anthropic ships the publicly available member of its top-tier Mythos lineup — state-of-the-art at coding, research and agentic work — live the same day on claude.ai, the API, AWS Bedrock and Google Vertex AI.

80.3%

SWE-Bench Pro score

Token context, standard pricing

94%+

Market odds of a June 9 launch

SWE-Bench Pro — coding performance

Higher columns mean stronger results on the same benchmark.

80.3%

Fable 5

69.2%

Opus 4.8

58.6%

GPT-5.5

API pricing vs Opus

Per million tokens (MTok). Fable 5 input is 2× Opus.

INPUT — $/MTok

Opus

$10

Fable 5

OUTPUT — $/MTok

$25

Batch

$50

Standard

Prompt caching can discount input up to 90%; better token efficiency can lower overall task cost.

Built-in safety routing

High-risk queries — cybersecurity or biology — are automatically routed to Opus 4.8.

<5%

Of queries trigger rerouting

Invite-only

Top-tier Mythos 5 stays restricted

Adaptive

Thinking always on; Extended off

WHAT'S LANDING WELL

Rated top-class for coding and complex reasoning
Beats Opus even at low/medium reasoning — using fewer tokens
Single prompts completing long-running work

THE FRICTION

Rate limits drained fast on the $20/mo plan (72% → 28%)
High price pushing heavy users to higher tiers
Some guardrails tightened vs the Mythos Preview

Continue reading

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