Anthropic's Sonnet 5.5 costs up to 30% less per task. For estimators, the test is bid leveling, not the benchmark score
Anthropic released Claude Sonnet 5.5 on September 28 at the same $2/$10 token price as Sonnet 5, with claims of faster output and lower cost per task. An estimating shop should price its own repeatable jobs, such as bid leveling and scope-gap checks, before switching models.
Anthropic released Claude Sonnet 5.5 on September 28 at the same $2 per million input tokens and $10 per million output tokens as Sonnet 5, and says it can finish a task for up to 30% less. For an estimating shop, that changes the math on repeatable back-office jobs like bid leveling and scope-gap checks, but only your own bid packages can tell you by how much.
What did Anthropic actually announce?
Sonnet 5.5 is the second model in the Claude 5.5 family. The reported details:
- Price: unchanged from Sonnet 5, $2 input and $10 output per million tokens, $0.20 per million for cache reads.
- Speed and cost: output more than 30% faster than Sonnet 5, and up to 30% lower cost per task, mostly because it uses fewer tokens and fewer tool calls, not because the sticker price fell.
- Positioning: Anthropic frames it as strongest at well-scoped everyday work, including bug fixes and polished documents, slides, and spreadsheets.
- Benchmark: 70.6% on Terminal-Bench 4.0, an agentic coding test, versus 10.3% for Sonnet 5. Coverage says it nearly matches Opus 5.5 on several benchmarks.
A coding benchmark tells an estimator very little directly. The useful part is the cost-per-task claim and the spreadsheet focus.
What does this mean for estimators?
Estimating has a lot of well-scoped, repeatable jobs. Those are the jobs where a cheaper, faster model is worth testing:
| Job | Why it fits a mid-tier model | What a person still owns |
|---|---|---|
| Bid leveling across sub quotes | Structured comparison in a spreadsheet | Inclusions, exclusions, alternates, qualifications |
| Scope-gap checklist against a spec section | Clear inputs, checkable output | Deciding what the gap costs |
| Addenda change summaries | Short, well-defined documents | Confirming nothing was missed in the drawings |
| Quote-to-template data entry | Repetitive formatting | Spot-checking quantities and units |
Open-ended judgment, like deciding how a sub's odd exclusion affects your number, is where reporting suggests staying on the larger model.
Should a mid-size estimating shop switch now?
Not on the announcement alone. Reported comparisons show Opus 5.5 at $4 input and $20 output per million tokens, so Sonnet 5.5 is half the list price. But one comparison found that at maximum effort, Sonnet 5.5 cost more per task than Opus 5.5 ($10.67 versus $8.40). Cost per task moves with effort setting and job type. A simple test:
- Pick three closed bids where you already know the right leveled answer.
- Run the same leveling job on both models at the effort levels you would actually use.
- Record the cost, the time, and every error against your known answer.
- Switch only the jobs where the cheaper model matched your answer.
If your firm reaches these models through a vendor's product instead of directly, none of this pricing reaches you automatically. Ask the vendor which model runs your jobs and whether a cheaper one changes your bill.
The takeaway
Treat the 30% figure as Anthropic's claim for its own tests. Your bid packages, with their scanned quotes and inconsistent sub formats, are a different workload. Price three closed bids on both models this week, and let the error count, not the benchmark, decide which one does your leveling.
- What did Anthropic release on September 28, 2026?
- Anthropic released Claude Sonnet 5.5, the second model in its Claude 5.5 family. It keeps Sonnet 5 pricing of $2 per million input tokens and $10 per million output tokens, and Anthropic says it generates output more than 30% faster and can cut the cost of a task by up to 30%.
- Is Sonnet 5.5 cheaper than Opus 5.5 for estimating work?
- On list price, yes: Opus 5.5 is reported at $4 input and $20 output per million tokens, twice Sonnet 5.5. Cost per task depends on how many tokens and tool calls a job uses, so an estimating firm should time and price its own bid-leveling or scope-check runs on both models.
- Can an AI model level subcontractor bids without an estimator checking it?
- No. Anthropic pitches Sonnet 5.5 for well-scoped tasks like spreadsheets and documents, but a leveled bid comparison still needs an estimator to confirm inclusions, exclusions, and alternates against the drawings and spec.
- Does Sonnet 5.5 have construction-specific features?
- Nothing in the launch coverage is construction-specific. Any estimating use comes from general strengths in spreadsheets, documents, and coding, so results on a firm's own bid packages need to be tested.