Claude Model Selection Guide
Anthropic's current Claude family covers four main workload profiles: demanding long-horizon work, complex engineering, everyday high-quality execution, and fast low-cost processing. Anthropic's official model overview recommends starting with Claude Opus 5.5 when you are unsure which model to choose, and escalating to Claude Fable 5.1 when Opus 5.5 still falls short at higher effort.
Current models
| Model | Positioning | Context | Max output | API input / output | Default effort |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Hardest reasoning, research, and long-horizon agents | 1M | 128K | $10 / $50 | high |
| Claude Opus 5.5 | Long-running agentic coding and knowledge work | 1M | 128K | $4 / $20 | medium |
| Claude Sonnet 5.5 | Best balance of speed and intelligence | 1M | 128K | $2 / $10 | high |
| Claude Haiku 4.5 | High-throughput, low-latency work | 200K | 64K | $1 / $5 | Not supported |
Prices are per million tokens. Cache reads, cache writes, Batch, and Fast mode are priced separately. Deployments through Bedrock, Google Cloud, and Microsoft Foundry use platform-specific model IDs.
Which model should you choose?
Start with Claude Opus 5.5
Opus 5.5 is Anthropic's current recommended starting point for most workloads. It is a strong fit for:
- changes across multiple directories or repositories;
- long-running Claude Code sessions;
- agents that need tool use, verification, and self-checking;
- research, financial analysis, documents, and complex business work;
- production tasks where reliability matters but Fable 5.1 pricing is difficult to justify.
Opus 5.5 always has adaptive thinking enabled. Use the effort parameter to control reasoning depth, latency, and cost. The default effort is medium; do not carry over the high default from Claude Opus 5 without re-evaluating it.
Escalate to Fable 5.1 when needed
Evaluate Fable 5.1 when Opus 5.5 still cannot complete a task reliably at higher effort. Fable 5.1 is intended for deeper reasoning, long-horizon planning, and cross-domain research. It is slower and more expensive, so it should not be the default for every agent turn.
Use Sonnet 5.5 for everyday work
Sonnet 5.5 is the cost-effective choice for routine coding, content generation, customer support, classification, and interactive products. It is faster and cheaper than Opus 5.5, making it a good default for high-volume calls with escalation for failed or high-risk tasks.
Use Haiku 4.5 for speed and volume
Haiku 4.5 is suited to fast responses, batch processing, routing, classification, and simple extraction. Its context and output limits are lower, so it should not be treated as a default replacement for long-running coding agents.
A practical routing strategy
Complex and costly-to-fail work -> Opus 5.5
Opus 5.5 still falls short -> Fable 5.1
High-volume everyday work -> Sonnet 5.5
Simple, fast, high-throughput -> Haiku 4.5
Track more than the model name: effort, tool calls, input and output tokens, latency, retries, and human correction time. The real optimization target is the total cost of an accepted result, not the price of one request.