Solve Hard Coding Tasks with Claude Fable 5 Adaptive Thinking

This example exercises Claude Fable 5 on hard coding tasks using the xhigh effort level. Claude Fable 5 is Anthropic's most powerful model, a new tier above Opus, and always uses adaptive thinking, so no thinking configuration is needed. The example evaluates distributed-systems debugging with incomplete information, production-quality code generation with concurrency concerns, and security-focused code review with prioritized feedback.

Use this example when you need to evaluate Fable 5 specifically on hard coding and agentic work at high or extra-high effort, where its adaptive thinking and large context matter. It is not appropriate for casual, low-effort tasks - effort defaults to high and xhigh is recommended for coding, and thinking consumes output tokens before any visible text, so a tight max_tokens budget can be spent entirely on thinking with an empty visible response.

Set ANTHROPIC_API_KEY, then scaffold and run:

npx promptfoo@latest init --example anthropic/fable-5-coding
cd fable-5-coding
export ANTHROPIC_API_KEY=your_api_key_here
npx promptfoo@latest eval
npx promptfoo@latest view

Fable 5 always thinks adaptively - unlike Opus 4.7/4.8 where adaptive thinking is opt-in - so promptfoo automatically converts a legacy thinking: { type: enabled, budget_tokens: N } config to adaptive and omits thinking: { type: disabled }, since the model rejects both. Thinking summaries are opt-in: by default the API omits thinking content (an empty thinking block, excluded from output by promptfoo), so set thinking: { type: adaptive, display: summarized } for a readable summary. Fable 5 rejects temperature, top_p, and top_k at the model level, and promptfoo omits them automatically. The model supports a 1M-token context and up to 128K output tokens, priced at $10/$50 per million input/output tokens.

Fable 5 is also reachable through AWS Bedrock (bedrock:global.anthropic.claude-fable-5 via Runtime/Converse, which requires opting the account into provider data sharing via aws bedrock put-account-data-retention --mode provider_data_share, or bedrock:messages:anthropic.claude-fable-5 for Bedrock's Anthropic-compatible Messages endpoint in us-east-1/eu-north-1), Google Vertex (vertex:claude-fable-5), and Azure AI Foundry (pointing anthropic:messages:claude-fable-5 at https://<resource>.services.ai.azure.com/anthropic via apiBaseUrl). Across all providers, promptfoo automatically omits unsupported sampling parameters and normalizes unsupported thinking configs for Fable 5; non-global Bedrock/Vertex regional endpoints carry a 10% price premium that promptfoo includes in cost calculations.

Teams evaluating Claude Fable 5 specifically for demanding coding and agentic tasks - distributed-systems debugging, concurrency-aware code generation, security-focused review - who need to understand its adaptive-thinking, sampling-parameter, and pricing differences from Opus before deploying it across AWS, Google, or Azure.