Breaking

Anthropic Debuts Top-Tier Claude Fable 5, Cutting Claude Code Prompt by 80%

July 2, 2026 at 15:15 EDT

Foundation Models
Software Dev & Coding
AI Agents

On June 9, 2026, Anthropic released "Claude Fable 5" (API model ID: claude-fable-5) and a restricted variant "Claude Mythos 5," introducing a new "Mythos-class" tier that sits above the previous Opus. Positioned for long-horizon agentic work, the model prompted an Anthropic engineer to reveal that Claude Code's system prompt had been trimmed by 80%.

June 9, 2026 · Anthropic

Claude Fable 5: Mythos-Class Power, Now Public and Guardrailed

Anthropic's first Mythos-class model released to everyone — state-of-the-art coding and agentic work with safety guardrails. An engineer says its sharper comprehension let them cut Claude Code's system prompt by 80%.

80%

shorter Claude Code system prompt — the model needs fewer instructions

token context window, with up to 128K output tokens

<5%

of sessions trigger guardrails — and they fall back, not refuse

Claude Code system prompt, before vs after Fable 5

Fewer instructions AND higher performance — a signal of the model's own understanding.

100%

Prior models

20%

With Fable 5

The Claude model tiers

Fable 5

Safe public release of Mythos-class — SOTA coding & agentic work

Public

Mythos 5

Same specs as Fable 5, guardrails removed

Partners

Opus 4.5

Prior high-performance tier

Public

Sonnet 5

Tier below Fable 5

Public

Praised

Coding & agentic work called a step-change
Stable sub-agent spawning, low context loss on long tasks
Makes previously impossible tasks feasible

Flagged

High cost and latency
Priced a tier above Opus 4.5 ($5/$25 per M tokens)
Restrictions on AI researchers have drawn pushback

Availability — live June 9, 2026

Rolled out across the Claude API, AWS Bedrock, Google Cloud, Microsoft Foundry and the Claude apps, with Snowflake Cortex AI integration the same day. Temporary export limits applied post-launch, before a global re-rollout around July 1, 2026.

Continue reading

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