Claude Sonnet 4.5vsClaude 3.5 Sonnet
Across 4 shared benchmarks, Claude Sonnet 4.5 leads overall: Claude Sonnet 4.5 wins 4, Claude 3.5 Sonnet wins 0, with 0 ties and an average score difference of +7.74.
Claude Sonnet 4.5
Anthropic · 2025-09-30 · AI model
Claude 3.5 Sonnet
Anthropic · 2024-06-21 · Multimodal model
Claude Sonnet 4.54 wins(100%)(0%)0 winsClaude 3.5 Sonnet
Benchmark scores
Grouped by capability, sorted by largest gap within each. 4 shared benchmarks.
General Knowledge
Claude Sonnet 4.5 2/2| Benchmark | Claude Sonnet 4.5 | Claude 3.5 Sonnet | Diff |
|---|---|---|---|
| GPQA Diamond | 73.7094 / 175 | 59.40138 / 175 | +14.30 |
| MMLU Pro | 885 / 124thinking | 77.6472 / 124 | +10.36 |
Math and Reasoning
Claude Sonnet 4.5 2/2| Benchmark | Claude Sonnet 4.5 | Claude 3.5 Sonnet | Diff |
|---|---|---|---|
| FrontierMath | 5.2038 / 60 | 152 / 60 | +4.20 |
| FrontierMath - Tier 4 | 2.1056 / 80Normal (No Tools) | 072 / 80Normal (No Tools) | +2.10 |
Specs
| Field | Claude Sonnet 4.5 | Claude 3.5 Sonnet |
|---|---|---|
| Publisher | Anthropic | Anthropic |
| Release date | 2025-09-30 | 2024-06-21 |
| Model type | AI model | Multimodal model |
| Architecture | Dense | Dense |
| Parameters | 0.0 | Not available |
| Context length | 1000K | 200K |
| Max output | 65536 | Not available |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | Claude Sonnet 4.5 | Claude 3.5 Sonnet |
|---|---|---|
| Text input | 3 美元/100 万tokens | Not public |
| Text output | 15 美元/100 万tokens | Not public |
| Cache read | 3.75 美元/100 万tokens | Not public |
| Cache write | 0.3 美元/100 万tokens | Not public |
One or both models have incomplete public pricing.
Summary
- Claude Sonnet 4.5leads in:General Knowledge (2/2), Math and Reasoning (2/2)
On average across the 4 shared benchmarks, Claude Sonnet 4.5 scores 7.74 higher.
Largest single-benchmark gap: GPQA Diamond — Claude Sonnet 4.5 73.70 vs Claude 3.5 Sonnet 59.40 (+14.30).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.