Labs

Anthropic lowers Fable
5.2 rates for
harder coding work

Lower token prices and claimed coding gains could make customer-acquisition experiments less costly—but small teams still need to test reliability, conversion and data handling.

By Casey VanderhoofFrontier labs correspondent·November 11, 2026

Anthropic released Claude Fable 5.2 on November 9, with standard API pricing of $8 per million input tokens and $40 per million output tokens. TechCrunch reports the input rate fell from $10, a 20% reduction.

For small-business owners, that suggests cheaper customer-acquisition experiments, not proven returns. One hypothetical: “a hypothetical project could improve a B2B inquiry form or consumer-facing booking flow, then test inquiry validity and completion rates.”

Coding gains, with caveats

Anthropic claims up to 30% faster output generation than Fable 5.1 and fewer tokens for challenging tasks at Medium effort, the default in Claude apps and Claude Code.

Its company-reported FrontierCode 2.0 testing puts high-effort Fable 5.2 at 45.3%, against Opus 5.5’s best 45.5%, at roughly half the cost per task. At Xhigh effort, Fable 5.2 reached 49.1%, versus Fable 5.1’s 32.4%.

On WANDR, which tests large data-collection tasks, Anthropic says Medium-effort Fable 5.2 matches Opus 5.5’s top score for about 40% of the cost per task. AutomationBench scores were 51.4% for Fable 5.2, 31.4% for Fable 5.1 and 42.5% for Opus 5.5. These are company-reported evaluations, not small-business acquisition campaigns.

Anthropic still recommends starting with Opus 5.5 for most coding and knowledge tasks, moving to Fable for more ambitious, open-ended work. It reports agents using an earlier version migrated a production service from Python to Rust; startup is now about 26 times faster.

Access and operating conditions

AWS says Fable 5.2 can work for hours or days unattended, solve engineering problems across repositories and test its work. Access is through Amazon Bedrock and Claude Platform on AWS. Google Cloud describes extended work on ambiguous problems and orchestration of subagents.

Google Cloud’s model listing marks claude-fable-5-2 generally available, with maximum input of 1 million tokens and output of 128,000 tokens. It accepts text, images and PDFs and produces text. Supported capabilities include web search, computer use, batch predictions, prompt caching and function calling. US, European and global endpoints are listed; retirement will be no sooner than November 9, 2027.

Anthropic’s optional fast mode promises up to 2.5 times faster generation, charging $16 per million input tokens and $80 per million output tokens. Google enables it with "speed": "fast". Doubled rates also cover cache writes and hits; web-search tool calls retain standard pricing.

Fast mode uses the separate anthropic-claude-fable-fast quota bucket. It doesn’t support Provisioned Throughput and always uses pay-as-you-go billing. Google says requests with the dedicated-throughput header receive an HTTP 429 error.

Google Cloud says prompts and responses are stored for up to 30 days solely for abuse monitoring; enabling the model requires consent to its Advanced AI Safety Addendum. AWS says covered models require safety-monitoring retention, though eligible customers can use zero data retention until Enterprise Frontier Safeguards becomes available.

Lower token rates don’t establish total project costs, permission to process customer data or acquisition results. A narrowly scoped lead-capture test still needs checks on quality, operating cost and conversion before a small team relies on it. The announcement doesn’t establish reliability without oversight.

Sources