Breaking

Anthropic Reroutes Fable 5 Frontier LLM Development Prompts to Opus 4.8

June 11, 2026 at 05:51 EDT

Foundation Models
Policy & Regulation
AI Agents

On June 11, 2026, Anthropic changed how its top-tier Claude Fable 5 model handles prompts related to "frontier LLM development," transparently rerouting such requests to the prior flagship, Claude Opus 4.8, as a fallback. The move follows community backlash, and users are notified when a request is flagged.

June 11, 2026 · Anthropic

Anthropic Walks Back Silent "Nerfing" in Claude Fable 5

After a developer backlash over undisclosed capability limits on "frontier LLM development" prompts, Anthropic calls it the "wrong tradeoff" — and will transparently route those requests to Claude Opus 4.8 instead of silently weakening them.

0.03%

of traffic affected by the silent restrictions

token context window for long-horizon agentic work

2×

the price of Opus 4.8 (input & output)

Fable 5 vs Opus 4.8 — Benchmarks

Higher score = better. Same unit (% pass), drawn to scale.

SWE-Bench Pro

80.3%

Fable 5

69.2%

Opus 4.8

FrontierCode Diamond

29.3%

Fable 5

13.4%

Opus 4.8

From Silent → Transparent

BEFORE

Prompt modification, steering vectors & PEFT capability limits — applied with no notice.

→

AFTER

Flagged requests transparently routed to Opus 4.8; API returns a refusal reason, server-side fallback to follow.

Praised

Endurance on long, multi-day agentic sessions
Code-review accuracy on complex projects
Transparency fix seen as a step forward

Criticized

Silently weakening a Mythos-class model called self-defeating
More false positives on cyber & bio queries
Distrust of centralized, hidden safety control

Continue reading

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