Define comparison criteria
The work assignment is to classify models by quality, speed, cost and task. Define purpose, owner and permitted operating boundary before the first test.
The domain review baseline covers a fixed test sample and currently verified model and price data. Assumptions and missing information remain visible in the result.
Compare using the same task
- Select a fixed sample of intended tasks and define domain-specific assessment criteria.
- Record available models and relevant pricing information from official sources with a review date.
- Test candidates with identical inputs and record quality, speed and cost assumptions.
- Explain the selection for the specific purpose and define when it should be reviewed again.
Document the selection rationale
A comparison record contains identical inputs for each candidate, expert assessment of answers, measured completion times and dated cost assumptions. The weighting of criteria is traceable.
Avoid: old model names or benchmark scores treated as permanent truth. Without current model and pricing data or a representative work sample, no reliable winner can be established for business use.
Model selection after Opus 5.5 and Sonnet 5.5
Opus 5.5 joined the Claude lineup on 22 September 2026, followed by Sonnet 5.5 on 28 September. A list limited to Fable 5.1, Opus 5, Sonnet 5 and Haiku 4.5 therefore no longer covers the current choice. Anthropic positions Sonnet 5.5 as a faster complement for well-defined tasks. API list prices remain the same as Sonnet 5: two US dollars per million input tokens and ten per million output tokens. Lower task costs are a vendor efficiency claim, not a further token-price cut.
Compare models actually available to you using identical tasks, recording accepted results, rework and total cost. Before switching API models, also check whether your parameters and tool calls remain compatible. The Opus 5.5 guide covers its evaluation and API migration; the Fable guide remains a separate assessment of that model. The AI news roundup of 29 September adds the Sonnet launch and other new capabilities.
Decision matrix
| Decision point | Proceed when | Stop when |
|---|---|---|
| Comparison baseline | Documented: a fixed test sample and currently verified model and price data. | Scope, data or accountability remains unresolved. |
| Task test | A comparison record contains identical inputs for each candidate, expert assessment of answers, measured completion times and dated cost assumptions. The weighting of criteria is traceable. | There is only an unevaluated demo without acceptance evidence. |
| Decision boundary | Owner, approval, fallback and next review date are defined. | Avoid: old model names or benchmark scores treated as permanent truth. Without current model and pricing data or a representative work sample, no reliable winner can be established for business use. |
Keep it verifiable
Primary sources
- Claude SonnetSource checked:
- Claude EnterpriseSource checked:
- Anthropic: commercial data and trainingSource checked:
- Anthropic: Remote MCP connectorsSource checked:
- Anthropic: Claude model overviewSource checked:
- Anthropic: Claude API pricingSource checked:
- Claude Fable 5.1 model overviewSource checked:
- Claude models overviewSource checked:
- Anthropic: Claude Opus 5.5Source checked:
- Anthropic: Claude Platform release notesSource checked:
- Anthropic: Introducing Claude Sonnet 5.5Source checked:
- Anthropic: Claude Sonnet 5.5 model overviewSource checked:
