Claude Haiku 5.5 vs Claude Sonnet 5.5: Which One Should You Use?
Anthropic's new Claude Haiku 5.5 is its fastest and cheapest model, while Claude Sonnet 5.5 is the stronger all-rounder. Here is how they compare on benchmarks, cost, and pricing.
Benchmarks
| Benchmark | Claude Haiku 5.5 | Claude Sonnet 5.5 | Difference |
|---|---|---|---|
| GDPval-AA v2.1 | 1620 Elo | 1840 Elo | Sonnet +220 Elo |
| AA-Briefcase v1.1 | 1578 Elo | 1824 Elo | Sonnet +246 Elo |
| OSWorld 2.1 (offline subset) | 72.4% | 83.9% | Sonnet +11.5 points |
| Humanity's Last Exam, no tools | 45.9% | 56.9% | Sonnet +11.0 points |
| Humanity's Last Exam, with tools | 57.4% | 64.5% | Sonnet +7.1 points |
| Terminal-Bench 4.0 | 39.2% | 70.6% | Sonnet +31.4 points |
| FrontierCode 1.1 (Main) | 46.4% | 52.1%Xhigh effort | Sonnet +5.7 points |
| Chartography, no tools | 46.4% | 61.6% | Sonnet +15.2 points |
Artificial Analysis benchmarks
| Benchmark | Claude Haiku 5.5 (Max) | Claude Sonnet 5.5 (Max) | Difference |
|---|---|---|---|
| Intelligence Index v4.3.2 | 43 | 56 | Sonnet +13 |
| AA-Briefcase v1.1 | 1578 Elo | 1823 Elo | Sonnet +245 Elo |
| GDPval-AA v2.1 | 1620 Elo | 1839 Elo | Sonnet +219 Elo |
| AutomationBench-AA | 35.4% | 71.8% | Sonnet +36.4 points |
| Terminal-Bench 4.0 | 32.8% | 63.6% | Sonnet +30.8 points |
| SciCode | 55.0% | 61.0% | Sonnet +6.0 points |
| Humanity's Last Exam | 44.4% | 55.0% | Sonnet +10.6 points |
| GDP.pdf | 20.8% | 25.8% | Sonnet +5.0 points |
| CritPt | 18.9% | 31.4% | Sonnet +12.5 points |
| AA-Omniscience | 11 | 32 | Sonnet +21 |
| AA-LCR v1.1 | 82.7% | 82.7% | Tie |
Cost and efficiency
| Metric | Claude Haiku 5.5 (Max) | Claude Sonnet 5.5 (Max) | Difference |
|---|---|---|---|
| Intelligence Index | 43 | 56 | Sonnet +13 |
| Cost per Intelligence Index task | $0.21 | $5.46 | Haiku about 96% cheaper |
| Cost to run the Intelligence Index | $330 | $7,260 | Haiku about 95% cheaper |
| Output tokens per task | 162K | 197K | Haiku uses about 18% fewer |
| Reasoning tokens per task | 129K | 147K | Haiku uses about 12% fewer |
| Output speed | 242 tokens/s | 129 tokens/s | Haiku about 1.9x faster |
| Context window | 1M tokens | 1M tokens | Same |
Pricing and specs
| Specification | Claude Haiku 5.5 | Claude Sonnet 5.5 | Difference |
|---|---|---|---|
| Input, prompts up to 100K | $0.10 | $2 | Haiku 95% cheaper |
| Input, prompts over 100K | $0.50 | $2 | Haiku 75% cheaper |
| Output, prompts up to 100K | $0.50 | $10 | Haiku 95% cheaper |
| Output, prompts over 100K | $2.50 | $10 | Haiku 75% cheaper |
| Cache read, prompts up to 100K | $0.01 | $0.10 | Haiku 90% cheaper |
| Cache read, prompts over 100K | $0.05 | $0.10 | Haiku 50% cheaper |
| Cache write, prompts up to 100K | $0.125 | $2.50 | Haiku 95% cheaper |
| Cache write, prompts over 100K | $0.625 | $2.50 | Haiku 75% cheaper |
| Context window | 1M tokens | 1M tokens | Same |
| Adjustable reasoning effort | Yes | Yes | Same |
| Text input | Yes | Yes | Same |
| Image input | Yes | Yes | Same |
| Text output | Yes | Yes | Same |