Claude Fable 5 Launches at $10/MTok: Worth the Premium?

What launched

On June 9, 2026, Anthropic released Claude Fable 5 at $10.00/MTok input and $50.00/MTok output. It is the most expensive model across all three providers by a wide margin.

Fable 5 also brings a 1,000,000 token context window. That is 5x larger than Opus 4-8.

The confidence scores

Fable 5 earns the highest or tied-highest confidence across four task types:

• -Reasoning: 98/100 (rank 1, tied with GPT 5.6 Sol at 96)
• -Code generation: 96/100 (rank 1)
• -Code review: 95/100 (rank 1, tied with Sonnet 4-6)
• -Creative: 95/100 (rank 1)

For reasoning tasks, Fable 5 is 2 points above the next best model. For classification, it scores just 82/100, ranking 5th overall.

The price premium in real numbers

For a code generation workflow at 2,000 input and 4,000 output tokens per call:

• -Fable 5: (2,000/1M x $10) + (4,000/1M x $50) = $0.220/call
• -Claude Opus 4.8: (2,000/1M x $5) + (4,000/1M x $25) = $0.110/call
• -Claude Sonnet 5: (2,000/1M x $3) + (4,000/1M x $15) = $0.066/call
• -Gemini 3.1 Pro: (2,000/1M x $2) + (4,000/1M x $12) = $0.052/call

At 5,000 calls per day, Fable 5 costs $1,100/day versus $330/day for Sonnet 5.

When Fable 5 makes sense

Fable 5 is built for workloads where quality matters more than cost. Complex reasoning, novel code generation, and creative tasks are its strengths.

For classification, extraction, and simple QA, cheaper models score higher at 1/40th the price. The key is routing the right tasks to the right model.

The batch discount matters here

Fable 5 offers a 50% batch discount, bringing input to $5.00 and output to $25.00. That matches Opus list pricing. For non-latency-sensitive workloads, batching cuts the premium entirely.

Anthropic also offers 90% cache read discounts. Cached Fable 5 input drops to $1.00/MTok. For workloads with high cache hit rates, the effective cost is far lower.

See how Fable 5 compares for your workload at /teardown, or start tracking with a free dashboard.