Across 12 comparable benchmarks, Claude Sonnet 4.5 has better reported results on 11 and worse on 1. 1 further shared result is listed below, each with the reason it is not counted.
Moderate evidence12 of 13 shared results countedHow we compare
Claude 3.5 Sonnet New is no longer available through its official API. API prices for this version are not shown.
Price basis: Listed rates vs standard rates. The price section below shows the applicable periods.
| Fact | Claude 3.5 Sonnet New | Claude Sonnet 4.5 |
|---|---|---|
| Context window | 200K tokens | 1M tokens |
| Input / 1M | — | $3.00 |
| Output / 1M | — | $15.00 |
| Blended (4:1) | — | $3.46 |
| Status | Retired | Deprecated |
| Released | October 22, 2024 | September 30, 2025 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 8 shared benchmarks grouped by capability on a 0–100 scale. How averages are built
Claude Sonnet 4.5 ahead on 3 of 3 counted
Claude Sonnet 4.5 ahead on 1 of 1 counted
No fixed-scale average — see the rows in Benchmark results
Claude Sonnet 4.5 ahead on 2 of 2 counted
Claude Sonnet 4.5 ahead on 1 of 1 counted
Split 1–1 of 2 counted
Claude Sonnet 4.5 ahead on 1 of 1 counted
No fixed-scale average — see the rows in Benchmark results
No shared comparable results for Computer & tool use, Research & search, Long context, Cybersecurity.
Published scores, with matching modes where available. Gaps show Claude 3.5 Sonnet New relative to Claude Sonnet 4.5. Reading the results
| Benchmark | Claude 3.5 Sonnet New | Claude Sonnet 4.5 | Gap |
|---|---|---|---|
| GSOMaintenance & Optimization · OPT@1 (%) | 4.6Normal · tools | 14.7Normal · tools | −10.1 pts#20 vs #15 of 24 |
| SWE-bench VerifiedRepository Engineering · Accuracy | 49.0Normal · no tools | 82.0≠Parallel Thinking · tools | −33.0 pts#94 vs #8 of 108 |
| WeirdML v2ML Engineering · Average accuracy across 17 tasks (%) | 40.0Normal · tools | 47.7≠Thinking · tools · 16K | −7.7 pts#34 vs #27 of 39 |
| SimpleBenchCommonsense · Score (AVG@5) | 41.4Normal · no tools | 54.3Normal · no tools | 0.76×#75 vs #58 of 106 |
| FrontierMathMathematics · Accuracy | 2.1Normal · no tools | 5.2Normal · no tools | −3.1 pts#36 vs #29 of 48 |
| FrontierMath - Tier 4Mathematics · Accuracy | 0.0Normal · no tools | 4.2≠Thinking · no tools · 32K | −4.2 pts#44 vs #21 of 50 |
| GPQA DiamondScientific Reasoning · Accuracy | 65.0Normal · no tools | 73.7Normal · no tools | −8.7 pts#157 vs #124 of 200 |
| GeoBench ACWVisual Understanding · ACW country accuracy (%) | 62.0Normal · no tools | 60.0≠Thinking · no tools | +2.0 pts#26 vs #28 of 32 |
| VPCTVisual Understanding · Accuracy (%) | 33.0Normal · no tools | 39.8≠Thinking · no tools · 32K | −6.8 pts#24 vs #16 of 26 |
| Creative WritingWriting | 1451Normal · no tools | 1677Normal · no tools | 0.87×#66 vs #36 of 121 |
| HLEKnowledge Exams · Accuracy | 4.1Reported best | 33.6≠Thinking · tools | −29.5 pts#149 vs #84 of 151 |
| MMLU-ProKnowledge Exams · Accuracy | 78.0Normal · no tools | 88.0≠Thinking · no tools | −10.0 pts#72 vs #7 of 153 |
| METR Time Horizons v1.1Capability Frontier Metrics · 50% task-completion time horizon (minutes)Not counted · Speed and cost telemetry is reported, not scored | 29.6Normal · tools | 122≠Thinking · tools · 16K | 0.24×#17 vs #10 of 19 |
Maintenance & Optimization · OPT@1 (%)
Repository Engineering · Accuracy
ML Engineering · Average accuracy across 17 tasks (%)
Commonsense · Score (AVG@5)
Mathematics · Accuracy
Mathematics · Accuracy
Scientific Reasoning · Accuracy
Visual Understanding · ACW country accuracy (%)
Visual Understanding · Accuracy (%)
Writing
A dash means no result is available.
| Benchmark | Claude 3.5 Sonnet New | Claude Sonnet 4.5 |
|---|---|---|
| AIME 2024 | 16.0 | — |
| AIME2025 | — | 100.0 |
| CritPt | — | 1.1 |
| FrontierMath Tier 4 v2 | — | 2.4 |
| FrontierMath v2 | — | 23.9 |
| MATH | 78.3 | — |
| MATH-500 | 78.0 | — |
| MMLU | 89.3 | — |
| SciCode | — | 45.7 |
| Aider-Polyglot | 51.6 | — |
| APEX-SWE | — | 31.0 |
| CodeClash | — | 1389 |
| HumanEval | 93.7 | — |
| LiveCodeBench | 38.1 | — |
| SWE-bench | 29.4 | — |
| SWE-Bench Pro - Public | — | 43.6 |
| Terminal Bench 2.0 | — | 42.8 |
| Terminal Bench Hard | — | 35.6 |
| Terminal-Bench 1.0 | — | 50.0 |
| Vibe Code Bench v1.1 | — | 22.6 |
| Claw Bench | — | 88.1 |
| MCP-Atlas | — | 59.5 |
| OSWorld-Verified | — | 62.9 |
| Pinch Bench | — | 88.2 |
| ECI | — | 147 |
| LiveBench | — | 68.2 |
| Text Arena (Coding) | — | 1391 |
| IF Bench | — | 57.3 |
| EBR-bench | — | 2.4 |
| GDP.pdf | — | 5.2 |
| MMMU | — | 77.8 |
| MMMU-Pro | — | 68.9 |
| GDPval-AA | — | 39.0 |
| SAGE | — | 36.1 |
| τ²-Bench | — | 84.7 |
| τ²-Bench - Telecom | — | 98.0 |
| τ³-Banking | — | 24.5 |
| ARC-AGI-1 | — | 63.7 |
| ARC-AGI-2 | — | 13.6 |
| BALROG | 32.6 | — |
| BBH | 92.6 | — |
| Vending-Bench 2 | — | 3839 |
| BrowseComp | — | 24.1 |
| GAIA | — | 74.5 |
| SimpleQA | 28.4 | — |
| MedCode | — | 44.1 |
| MedScribe | — | 84.5 |
Time-of-day providers use peak-hour rates; fixed-price providers use standard rates. Off-peak discounts remain in the published details.
Drag the handle above the bar for cache hits, the one below it for the input : output mix · 5% steps
Blended price = effective input rate at 90% cache hits × 80% + output rate × 20%
A blended cost needs a published input and output price on both sides, and one of these two does not have them yet.
Claude 3.5 Sonnet New: Listed text rates; pricing tier not confirmed.
Claude Sonnet 4.5: Standard pricing. Anthropic official pricing (platform.claude.com/docs/en/about-claude/pricing) verified 2026-11-13; Deprecated 2026-09-30; tentative retirement 2026-11-30 (replacement: claude-sonnet-5-5)
| Price / 1M tokens | Claude 3.5 Sonnet New | Claude Sonnet 4.5 | Ratio |
|---|---|---|---|
| Input | — | $3.00 | — |
| Output | — | $15.00 | — |
| Cached input | — | $0.30 | — |
| Cache write | — | — | — |
| Blended | — | $3.46 | — |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
| Attribute | Claude 3.5 Sonnet New | Claude Sonnet 4.5 |
|---|---|---|
| Availability | Retired | Deprecated |
| Provider | Anthropic | Anthropic |
| Model type | Chat model | Chat model |
| First release | October 22, 2024 | September 30, 2025 |
| Context window | 200K tokens | 1M tokens |
| Max output | — | 64K tokens |
| Architecture | Undisclosed | Undisclosed |
| Documented capabilities | Multilingual | Reasoning, Multilingual |
| Input modalities | Text, Image | Text, Image |
| Output modalities | Text | Text |
| Weights | Closed / API only | Closed / API only |
| Commercial use | Proprietary | Proprietary |
These are reported-score comparisons, not controlled head-to-head tests.
What people found comparing these models, or using either one. Comments also appear on the models' own pages.
Used Claude 3.5 Sonnet New or Claude Sonnet 4.5? Sign in with Google to share what worked, what did not, and which API or setup you tested.
No community comments on either model yet.
Alternatives with the most benchmark overlap for this matchup.