Claude Opus 5.5 vs Sonnet 5.5: Pricing and API Rates
A technical comparison of Claude Opus 5.5 versus Sonnet 5.5 covering official pricing, prompt caching write and read rates, and published performance trade-offs.
Claude Sonnet 5.5 costs $2.00 per million input tokens and $10.00 per million output tokens, cutting the standard API token bill exactly in half compared to Claude Opus 5.5 at $4.00 input and $20.00 output. However, both frontier models share the exact same prompt cache read rate of $0.20 per million tokens, meaning cache-heavy agent pipelines see a much smaller price gap than standard API rates suggest.123
Architectural Positioning: How Anthropic Segments the 5.5 Tier
Anthropic launched Claude Opus 5.5 on September 22, 2026, targeting long-running agentic coding and knowledge work. Six days later, on September 28, 2026, Anthropic released Claude Sonnet 5.5 as a faster, lower-cost complement built for everyday scoped coding and tool-based work.145
Both models support text and image input, tool use, and 128K maximum output tokens out of a default 1M token context window. Anthropic reports that Sonnet 5.5 generates output over 30 percent faster than Sonnet 5 and costs up to 30 percent less per task, whereas Opus 5.5 is designed to sustain deeper judgment across complex, multi-step problem spaces.126
API Pricing Breakdown: Standard Tokens, Prompt Caching, and Writes
When comparing raw input and output tokens, Opus 5.5 costs exactly 2.0x what Sonnet 5.5 charges. On prompt caching, the write cost also scales at a 2.0x ratio ($2.50 per million tokens for Sonnet 5.5 versus $5.00 per million tokens for Opus 5.5). However, reading from cached prompt prefixes costs $0.20 per million tokens on both tiers.23
| Billing Dimension | Claude Sonnet 5.5 (claude-sonnet-5-5) | Claude Opus 5.5 (claude-opus-5-5) | Multiplier (Opus vs Sonnet) |
|---|---|---|---|
| Standard Input (per 1M tokens) | $2.00 | $4.00 | 2.0x |
| Standard Output (per 1M tokens) | $10.00 | $20.00 | 2.0x |
| Prompt Cache Write (per 1M tokens) | $2.50 | $5.00 | 2.0x |
| Prompt Cache Read (per 1M tokens) | $0.20 | $0.20 | 1.0x |
For a conversation turn with 100k input tokens and 100k output tokens, the approximate usage cost is $1.20 on Claude Sonnet 5.5 compared to $2.40 on Claude Opus 5.5. Because cached prompt reads remain flat at $0.20 per million tokens regardless of model tier, long multi-turn agent systems that maintain large static prefixes will observe their realized cost ratio drop well below 2.0x.23
Vendor Benchmark Comparison: SWE-Bench, Terminal-Bench, and Reasoning
Vendor evaluations show a split between full-repository engineering and terminal execution. On SWE-Bench Pro, Claude Opus 5.5 leads at 89.9% versus 81.3% for Sonnet 5.5. Opus 5.5 also leads on FrontierCode v1.1 at 54.4% compared to 46.2% for Sonnet 5.5.5
On Artificial Analysis Intelligence Index v4.1.1, Claude Opus 5.5 posts an intelligence score of 58.0 at maximum reasoning effort, compared to 56.0 for Claude Sonnet 5.5. Across everyday execution benchmarks, the margin narrows or flips: on Terminal-Bench 4.0, Sonnet 5.5 scores 70.6% versus 66.4% for Opus 5.5 at extra-high effort, while Sonnet 5.5 achieves 44.7% on AutomationBench compared to 42.5% for Opus 5.5.25
| Benchmark | Claude Sonnet 5.5 | Claude Opus 5.5 | Leader Margin |
|---|---|---|---|
| SWE-Bench Pro | 81.3% | 89.9% | Opus +8.6 |
| FrontierCode v1.1 | 46.2% | 54.4% | Opus +8.2 |
| Humanity's Last Exam (with tools) | 64.5% | 67.7% | Opus +3.2 |
| CursorBench 4.0 | 55.5% | 57.8% | Opus +2.3 |
| OSWorld 2.1 | 80.1% | 81.8% | Opus +1.7 |
| GDPval-AA v2.1 (Elo) | 1844 | 1846 | tie |
| AutomationBench | 44.7% | 42.5% | Sonnet +2.2 |
| Terminal-Bench 4.0 | 70.6% | 66.4% | Sonnet +4.2 |
Operational Trade-Offs: Latency, Thinking Modes, and Tool Controls
Thinking configurations alter latency and request execution on each model tier. In the official API documentation, Claude Opus 5.5 enforces always-on adaptive thinking with medium effort by default, and passing disabled to thinking returns a 400 error. Sonnet 5.5 defaults to high effort on the Claude Platform, but developers can bypass up-front thinking before tool execution by setting thinking to between_tools at high effort or below.14
For integration compatibility, both models return a 400 error if forced tool choice (tool_choice types any or tool) is supplied, requiring migration to auto with strict tool execution instead.45
Workload Decision Framework: When to Route to Opus 5.5
Assign bug fixes, terminal command execution, and repeatable tool loops to Claude Sonnet 5.5 as the default route. At medium effort, Sonnet 5.5 exceeds Sonnet 5's top score for less than a tenth of the cost per task by executing terminal commands directly rather than spawning unnecessary subagents. In a vendor test of isolated unit repair, Sonnet 5.5 completed the work in 48 seconds for $0.19, compared to 68 seconds and $0.33 for Opus 5.5.57
Escalate to Claude Opus 5.5 when an open-ended code change spans multiple repository packages or when previous debug attempts produced regressions. On SWE-Bench Pro, Opus 5.5 achieves an 89.9% pass rate against 81.3% for Sonnet 5.5. Paying $4.00 per million input tokens and $20.00 per million output tokens is cost-effective when the 8.6-point margin on full repository changes resolves a task that would otherwise require repeated manual patching.235
Questions
Should I use Claude Opus 5 or Sonnet 5?
For current production deployments, Anthropic recommends Claude Sonnet 5.5 over earlier generation models for everyday coding, documents, and standard tool use. Claude Opus 5.5 should be used when tackling complex, open-ended reasoning tasks or repo-scale architecture problems that require sustained judgment.
What is the price difference between Claude Opus 5 and Sonnet 5?
In the 5.5 model family, standard input and output token rates on Opus 5.5 ($4.00 input, $20.00 output per million tokens) cost exactly twice as much as Sonnet 5.5 ($2.00 input, $10.00 output). However, both models bill prompt cache reads at an identical rate of $0.20 per million tokens.
Why is Claude Sonnet 5 so expensive?
Claude Sonnet 5.5 is priced at $2.00 per million input tokens and $10.00 per million output tokens, matching Sonnet 5 list pricing while offering higher throughput. When compared to lightweight models like Haiku 4.5 ($1.00 input, $5.00 output), Sonnet carries higher compute requirements to support its 128K token maximum output and advanced agentic capabilities.
Is Sonnet 5 better?
Claude Sonnet 5.5 outperforms Sonnet 5 across major benchmarks, including a jump from 63.2% to 81.3% on SWE-Bench Pro and from 10.3% to 70.6% on Terminal-Bench 4.0. Compared to Opus 5.5, Sonnet 5.5 offers lower latency and lower token cost while closely matching performance on everyday tool and terminal tasks.