Across 14 comparable benchmarks, Gemini 2.5 Pro Experimental 03-25 has better reported results on 11 and worse on 3.
Strong evidence14 of 14 shared results countedHow we compare
Gemini 2.5 Pro Experimental 03-25 is no longer available through its official API. API prices for this version are not shown.
Claude Sonnet 4 is no longer available through its official API. API prices for this version are not shown.
Price basis: Listed rates vs listed rates. The price section below shows the applicable periods.
| Fact | Gemini 2.5 Pro Experimental 03-25 | Claude Sonnet 4 |
|---|---|---|
| Context window | 1M tokens | 200K tokens |
| Input / 1M | — | — |
| Output / 1M | — | — |
| Blended (4:1) | — | — |
| Status | Retired | Retired |
| Released | March 25, 2025 | May 23, 2025 |
Blended is 4:1 input to output with 90% cache hits at listed rates — set your own mix.
Mean of 10 shared benchmarks grouped by capability on a 0–100 scale. How averages are built
Gemini 2.5 Pro Experimental 03-25 ahead on 2 of 3 counted
Gemini 2.5 Pro Experimental 03-25 ahead on 1 of 1 counted
No fixed-scale average — see the rows in Benchmark results
Gemini 2.5 Pro Experimental 03-25 ahead on 3 of 3 counted
Gemini 2.5 Pro Experimental 03-25 ahead on 1 of 1 counted
Split 1–1 of 2 counted
Gemini 2.5 Pro Experimental 03-25 ahead on 1 of 1 counted
Gemini 2.5 Pro Experimental 03-25 ahead on 1 of 1 counted
Claude Sonnet 4 ahead on 1 of 1 counted
No fixed-scale average — see the rows in Benchmark results
No shared comparable results for Research & search, Cybersecurity.
Published scores, with matching modes where available. Gaps show Gemini 2.5 Pro Experimental 03-25 relative to Claude Sonnet 4. Reading the results
| Benchmark | Gemini 2.5 Pro Experimental 03-25 | Claude Sonnet 4 | Gap |
|---|---|---|---|
| Aider-PolyglotAgentic Development · Accuracy | 72.9Normal · no tools | 61.3≠Thinking · no tools · 32K | +11.6 pts#8 vs #14 of 47 |
| LiveCodeBenchAlgorithmic Coding · Pass @K | 70.4Normal · no tools | 48.5Normal · no tools | 1.45×#55 vs #88 of 113 |
| SWE-bench VerifiedRepository Engineering · Accuracy | 63.8Normal · no tools | 80.2≠Parallel Thinking · tools | −16.4 pts#75 vs #15 of 108 |
| SimpleBenchCommonsense · Score (AVG@5) | 51.6Normal · no tools | 45.5≠Thinking · no tools | 1.13×#63 vs #73 of 106 |
| AIME 2024Mathematics · Accuracy | 92.0Normal · no tools | 43.4Normal · no tools | +48.6 pts#8 vs #45 of 54 |
| AIME2025Mathematics · Accuracy | 86.9Normal · no tools | 85.0≠Deep Thinking · tools | +1.9 pts#43 vs #48 of 86 |
| FrontierMath - Tier 4Mathematics · Accuracy | 4.2Normal · no tools | 0.0Normal · no tools | +4.2 pts#21 vs #44 of 50 |
| GPQA DiamondScientific Reasoning · Accuracy | 84.0Normal · no tools | 83.8≠Deep Thinking · tools | +0.2 pts#88 vs #91 of 200 |
| Claw BenchTool Orchestration · Accuracy | 80.4Thinking · tools | 77.8Thinking · tools | +2.6 pts#20 vs #23 of 29 |
| Pinch BenchTool Orchestration · Accuracy | 71.9Thinking · tools | 80.5Thinking · tools | −8.6 pts#30 vs #23 of 38 |
| Fiction.liveBenchLong Reasoning · 16k-token score (%) | 66.7Normal · no tools | 46.9Normal · no tools | +19.8 pts#15 vs #24 of 28 |
| VPCTVisual Understanding · Accuracy (%) | 48.0Normal · no tools | 34.0≠Thinking · no tools · 32K | +14.0 pts#8 vs #22 of 26 |
| Creative WritingWriting | 1396Normal · no tools | 1483Normal · no tools | 0.94×#76 vs #62 of 121 |
| HLEKnowledge Exams · Accuracy | 18.8Normal · no tools | 9.6≠Thinking · no tools | +9.2 pts#108 vs #130 of 151 |
Agentic Development · Accuracy
Algorithmic Coding · Pass @K
Repository Engineering · Accuracy
Commonsense · Score (AVG@5)
Mathematics · Accuracy
Mathematics · Accuracy
Mathematics · Accuracy
Scientific Reasoning · Accuracy
Tool Orchestration · Accuracy
Tool Orchestration · Accuracy
Long Reasoning · 16k-token score (%)
Visual Understanding · Accuracy (%)
Writing
Knowledge Exams · Accuracy
A dash means no result is available.
| Benchmark | Gemini 2.5 Pro Experimental 03-25 | Claude Sonnet 4 |
|---|---|---|
| CritPt | — | 1.1 |
| FrontierMath | — | 4.1 |
| IMO-ProofBench | — | 27.1 |
| IMO-ProofBench Advanced | — | 4.8 |
| MATH-500 | — | 99.1 |
| MMLU-Pro | — | 84.0 |
| CodeClash | — | 1223 |
| IOI 2024 (Vals v1) | — | 9.7 |
| IOI 2025 (Vals v1) | — | 4 |
| SWE-Bench Pro - Commercial | — | 9.1 |
| SWE-Bench Pro - Public | — | 42.7 |
| Terminal Bench Hard | — | 31.1 |
| Terminal-Bench 1.0 | — | 41.3 |
| OSWorld-Verified | — | 42.2 |
| PinchBench v2 | — | 48.8 |
| ECI | — | 142 |
| LiveBench | — | 61.3 |
| IF Bench | — | 55.0 |
| GeoBench ACW | 81.0 | — |
| MMMU | — | 74.4 |
| MMMU-Pro | — | 62.4 |
| GDPval-AA | — | 33.0 |
| SAGE | — | 35.0 |
| τ²-Bench | — | 52.0 |
| τ³-Banking | — | 16.7 |
| ARC-AGI-1 | — | 40.0 |
| ARC-AGI-2 | — | 5.9 |
| BALROG | 43.3 | — |
| SimpleQA | 52.9 | — |
| MedCode | — | 35.0 |
| MedScribe | — | 72.4 |
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.
Gemini 2.5 Pro Experimental 03-25: Listed text rates; pricing tier not confirmed.
Claude Sonnet 4: Listed text rates; pricing tier not confirmed.
| Price / 1M tokens | Gemini 2.5 Pro Experimental 03-25 | Claude Sonnet 4 | Ratio |
|---|---|---|---|
| Input | — | — | — |
| Output | — | — | — |
| Cached input | — | — | — |
| Cache write | — | — | — |
| Blended | — | — | — |
Estimates use the selected token mix. Cache writes, tool fees and long-context surcharges are excluded.
Lower prices are highlighted. The cache hit rate applies to input tokens only; models without a published cached-input rate use their normal input price for those tokens.Effective dates: Gemini 2.5 Pro Experimental 03-25 —; Claude Sonnet 4 2026-11-14.Reported cost per task, paired with the score from the same run. Settings and sources may differ; costs do not affect the verdict.
| Attribute | Gemini 2.5 Pro Experimental 03-25 | Claude Sonnet 4 |
|---|---|---|
| Availability | Retired | Retired |
| Provider | Google Deep Mind | Anthropic |
| Model type | Reasoning model | Reasoning model |
| First release | March 25, 2025 | May 23, 2025 |
| Context window | 1M tokens | 200K tokens |
| Max output | 64K tokens | 64K tokens |
| Architecture | Undisclosed | Undisclosed |
| Documented capabilities | Reasoning, Multilingual | Reasoning, Multilingual |
| Input modalities | Text, Image, Audio, Video | 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 Gemini 2.5 Pro Experimental 03-25 or Claude Sonnet 4? 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.