Breaking

Anthropic Redeploys Claude Fable 5, 3D Generation in Focus

July 4, 2026 at 15:32 EDT

Foundation Models
AI Agents
Media Generation

Anthropic redeployed its frontier model Claude Fable 5 on July 1, 2026 with updated safeguards, drawing attention for its ability to handle complex 3D generation and software development from a single prompt. The discussion was sparked after developer Peter Gostev tested the model on more than 60 of the hardest 3D generation tasks—modern cities, world wonders, art and more—and described the output quality as being in a league of its own.

July 1, 2026 · Anthropic

Claude Fable 5 Returns Worldwide — and 3D Generation Steals the Show

After a brief export-control suspension, Anthropic's frontier public model is back. A batch of 60+ high-difficulty one-shot 3D demos has convinced developers that AI 3D modeling has reached a practical stage.

60+

hard 3D generation tasks run post-redeployment

token context window

1-shot

a single prompt often yields usable output

Pricing: roughly double Opus 4.8

Output token cost, per 1M tokens (2× column ≈ 2× the price)

$25

Opus 4.8

$50

Fable 5

Input $10 · Output $50 per 1M tokens · Max output 128K tokens · Knowledge cutoff Jan 2026

How the public model handles risky prompts

User query →
Fable 5

→

Safety
classifiers

→

Cyber / bio flags
routed to Opus 4.8

Mythos 5 (restricted, via Project Glasswing) runs relaxed safeguards; Fable 5 ships to everyone with strict classifiers.

STRENGTHS

Top score on Cognition's FrontierCode
Leads the Hebbia Finance Benchmark
Beats Opus 4.8 on agentic coding
Stripe & GitHub: months of work in days
One-shot 3D / CAD, sims, refactoring

LIMITATIONS

Classifiers over-trigger on benign queries
One benchmark: ~1/3 of responses blocked
High price, slower speed, heavy token use
Public build may hold back full capability

Continue reading

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