Breaking

Anthropic Pulls Claude Fable 5 From Flat-Rate Subscriptions, Shifting Its Top Reasoning Model to Usage-Based Pricing

July 13, 2026 at 09:15 EDT

Foundation Models
AI Agents
Sales & Business

Anthropic's most advanced reasoning model, Claude Fable 5, is being removed from the flat-rate allowances of its Pro, Max, Team and some Enterprise plans, leaving pay-as-you-go usage credits as the only route to access it going forward. The change, described by the company as a temporary measure driven by extraordinary demand, effectively ends the bundled subscription access that many users had come to rely on since the model's launch.

July 12–13, 2026 · Anthropic

Claude Fable 5 Leaves Flat-Rate Plans — Pay-Per-Token Only

Anthropic's most advanced reasoning model is being pulled from the bundled allowances of Pro, Max, Team and some Enterprise plans. Access continues only through usage credits. The company calls it a temporary, reversible move driven by extraordinary demand.

$10

per million input tokens (API rate)

$50

per million output tokens (API rate)

<5%

of sessions trigger safety fallback to Opus 4.8

Output Costs 5× More Than Input

Per million tokens under the usage-credit model — the same unit, drawn to scale.

$10

Input

$50

Output

The Timeline

Bundled inclusion was repeatedly extended before the final cutoff.

Jun 9

Fable 5 launches

→

Jun 22 → Jul 7

Inclusion extended

→

Jul 12–13

Removed from flat rate

Keep Using It

Pay per token via usage credits, or run the "principal engineer" trick — have Fable 5 audit a repo into 10–16 skill files and hand them to cheaper Opus/Sonnet models.

The Friction

High per-token cost, capacity constraints from concentrated demand, and occasional false positives from safety classifiers on benign coding. Some subscribers are reconsidering their plans.

Skill-transfer cost: generating the reusable manuals can consume roughly 30% of a weekly allowance — but practitioners consider it worthwhile to preserve Fable 5's judgment at lower cost.

Continue reading

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