Claude Haiku 5.5 vs Claude Sonnet 5.5: Specs & Benchmark Comparison
Claude Haiku 5.5 and Claude Sonnet 5.5 both come from Anthropic, Claude Sonnet 5.5 was released in September 2026, and Claude Haiku 5.5 followed a month later in October 2026.
The two models share 36 published benchmarks. Claude Sonnet 5.5 leads on 34 of them, Claude Haiku 5.5 on 2. The widest gaps are on Terminal-Bench-Science 0.1, where Claude Sonnet 5.5 scores 59.9% against 20.6%; Terminal-Bench 4.0, where Claude Sonnet 5.5 scores 70.6% against 39.2%. Averaged across everything we track, Claude Haiku 5.5 sits at 57.7% and Claude Sonnet 5.5 at 67.4%.
Claude Haiku 5.5 is the cheaper API at $0.10 per million input tokens and $0.50 per million output tokens, roughly 20 times cheaper than Claude Sonnet 5.5 at $2 and $10. Both accept a context window of about 1M tokens. Both accept text and images as input.
| Characteristic | Claude Haiku 5.5 | Claude Sonnet 5.5 |
|---|---|---|
| Company | Anthropic | Anthropic |
| Release Date | October 7, 2026 | September 28, 2026 |
| Parameters | — | — |
| Multimodal | Yes | Yes |
| Context (input) | 1.0M | 1.0M |
| Context (output) | 128K | 128K |
| Input Price / 1M | $0.10 | $2.00 |
| Output Price / 1M | $0.50 | $10.00 |
| Average Score | 57.7% | 67.4% |
| Benchmarks | ||
| Terminal-Bench-Science 0.1 | 20.6% | 59.9% |
| Terminal-Bench 4.0 | 39.2% | 70.6% |
| Biomedical image analysis | 43.5% | 72.2% |
| SWE-Bench Multimodal | 30.7% | 54.3% |
| PhysicianBench | 43.0% | 63.2% |
| FrontierSWE V2 | 43.8% | 61.9% |
| Protein Design — Sequence Generation | 33.1% | 51.0% |
| SWE-Bench Pro | 64.8% | 81.3% |
| Chartography (no tools) | 46.4% | 61.6% |
| Humanity's Last Exam (no tools, text-only) | 45.9% | 56.9% |
| Protein Design — Library Ranking | 45.3% | 54.8% |
| Protocols Troubleshooting | 58.0% | 67.3% |
| BenchCAD (with Python tool) | 87.0% | 96.3% |
| AA-Briefcase v1.1 | 52.6% | 60.4% |
| OSWorld 2.1 (partial) | 72.4% | 80.1% |
| BenchCAD | 67.0% | 74.7% |
| GDPval-AA 2.1 | 54.0% | 61.5% |
| Medicinal Chemistry (ADME) | 58.0% | 65.3% |
| Humanity's Last Exam (with tools) | 57.4% | 64.5% |
| SWE-bench Multilingual | 83.7% | 90.3% |
| OSWorld 2.1 (strict) | 37.1% | 43.5% |
| HealthBench Professional (raw) | 71.0% | 77.1% |
| FrontierCode 1.1 (Extended) | 58.4% | 64.4% |
| FrontierCode 1.1 | 46.4% | 52.1% |
| OfficeQA Pro | 60.3% | 65.6% |
| LatchBio SpatialBench Verified | 67.7% | 72.5% |
| HealthBench Professional | 64.8% | 69.2% |
| Global-MMLU | 87.8% | 92.1% |
| Chartography | 86.2% | 90.2% |
| MILU | 87.6% | 91.6% |
| HealthBench | 61.6% | 65.4% |
| OfficeQA | 73.5% | 76.9% |
| LatchBio SingleCellBench | 56.2% | 59.1% |
| Protocols Understanding (V2) | 64.1% | 66.6% |
| Program Bench | 82.0% | 79.7% |
| Morphology-to-molecule matching | 25.4% | 25.0% |
Visual Benchmark Comparison
Verdict
Claude Haiku 5.5 leads in 1 out of 4 comparison categories.
Both models show comparable average scores: Claude Haiku 5.5 — 0.6, Claude Sonnet 5.5 — 0.7.
Claude Haiku 5.5 is 20.0x cheaper: input $0.10/1M vs $2.00/1M tokens.
Same context size: 1M tokens.
Both models were released around the same time: 10/7/2026 and 9/28/2026.
More About These Models
Related Comparisons
Frequently Asked Questions
Which is better for coding — Claude Haiku 5.5 or Claude Sonnet 5.5?
Which model is cheaper — Claude Haiku 5.5 or Claude Sonnet 5.5?
Which has a larger context window — Claude Haiku 5.5 or Claude Sonnet 5.5?
The Claude Haiku 5.5 and Claude Sonnet 5.5 comparison is updated for 2026. Data includes benchmark results, API pricing, context window size and other specifications. For more detailed information, visit the Claude Haiku 5.5 or Claude Sonnet 5.5 page. See also the complete list of AI model comparisons.