Claude Haiku 4.5 vs Gemini 3.1 Pro Preview
Updated October 21, 2026. 82 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick Claude Haiku 4.5 for coding: it scores 49.8 on the Vibe Coding Index against 35.3 for Gemini 3.1 Pro Preview.
Pick Gemini 3.1 Pro Preview when the context window is larger, 1.05M against 200K tokens, or when what matters is LMArena WebDev or AA Coding Index, where it leads.
- Shared results
- 82
- Claude Haiku 4.5 only
- 12
- Gemini 3.1 Pro Preview only
- 49
- Like-for-like categories
- 8 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, four public coding benchmarks
Claude Haiku 4.5
Claude Haiku 4.5 scores 49.8 against 35.3.
3 and 1 benchmarks behind the scores
Intelligence
AA Intelligence Index
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview leads, 29.7 against 16.9.
Same board, both measured
Agentic work
AA Agentic Index
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview leads, 8.2 against 8.0.
Same board, both measured
Reasoning
GPQA Diamond
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview leads, 95.5% against 72.2%.
Same board, both measured
Math
AIME
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview leads, 98.1% against 82.7%.
Same board, both measured
Knowledge
MMLU Pro
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview leads, 91.0% against 78.7%.
Same board, both measured
Web design
Design Arena: websites
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview leads, 1234 against 1122.
Same board, both measured
Long documents
Listed context window
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview takes 1.05M tokens against 200K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Claude Haiku 4.5
Claude Haiku 4.5 $0.0035, Gemini 3.1 Pro Preview $0.0080 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
Claude Haiku 4.5
Claude Haiku 4.5 $0.56, Gemini 3.1 Pro Preview $0.99.
Published attempt cost and pass rate
At a glance
- Coding index
- Claude Haiku 4.5 49.8 from 3 benchmarks, Gemini 3.1 Pro Preview 35.3 from 1 benchmark.
- Widest category
- Reasoning, 31.1 points apart on average over 9 shared benchmarks.
- Widest single gap
- ARC-AGI-2: Claude Haiku 4.5 4.0% against Gemini 3.1 Pro Preview 77.1%, 73.1 points apart on a 0 to 100 scale.
- Output price
- Claude Haiku 4.5 $5.00 against Gemini 3.1 Pro Preview $12.00 per million output tokens, 2.4x apart.
- Context window
- Claude Haiku 4.5 takes 200K tokens, Gemini 3.1 Pro Preview 1.05M.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Claude Haiku 4.5 | Bars | Gemini 3.1 Pro Preview | Ties |
|---|---|---|---|---|
| Coding | 0 | 13 | 1 tie | |
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 8 | ||
| Reasoning | 0 | 9 | ||
| Math | 0 | 4 | ||
| Knowledge | 0 | 13 | ||
| Web design | 0 | 8 | ||
| General | 2 | 23 | ||
| All shared boards | 2 | 79 | 1 tie |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index013
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| SWE-bench Verified | 66.6% | Not measured | Only Claude Haiku 4.5 measured |
| SWE-bench Multilingual | 64.7% | Not measured | Only Claude Haiku 4.5 measured |
| LiveBench coding | 48.1% | Not measured | Only Claude Haiku 4.5 measured |
| LMArena WebDev | 1330 | 1447 | Gemini 3.1 Pro Preview |
| AA Coding Index | 43.9 | 68.8 | Gemini 3.1 Pro Preview |
| LiveCodeBench (Vals AI) | 41.2% | 88.5% | Gemini 3.1 Pro Preview |
| SWE-bench Verified (Vals AI) | 66.6% | 78.8% | Gemini 3.1 Pro Preview |
| Terminal-Bench 4 (Vals AI) | Not measured | 2.5% | Only Gemini 3.1 Pro Preview measured |
| Vibe Code Bench (Vals AI) | 11.4% | 32.0% | Gemini 3.1 Pro Preview |
| Code migration (Vals AI) | 7.5% | 17.3% | Gemini 3.1 Pro Preview |
| ProgramBench (Vals AI) | 0.0% | 0.0% | Tie |
| IOI (Vals AI) | Not measured | 51.8% | Only Gemini 3.1 Pro Preview measured |
| Terminal-Bench 2.1 (Vals AI) | 43.8% | 70.8% | Gemini 3.1 Pro Preview |
| Vibe Code Bench 1-100 (Vals AI) | Not measured | 6.7% | Only Gemini 3.1 Pro Preview measured |
| SciCode | Not measured | 58.7% | Only Gemini 3.1 Pro Preview measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 53.8% | Only Gemini 3.1 Pro Preview measured |
| Terminal-Bench | 35.5% | 80.2% | Gemini 3.1 Pro Preview |
| GSO-Bench | Not measured | 22.6% | Only Gemini 3.1 Pro Preview measured |
| WeirdML | 45.4% | 72.1% | Gemini 3.1 Pro Preview |
| MirrorCode | Not measured | 8.9% | Only Gemini 3.1 Pro Preview measured |
| ALE-Bench | Not measured | 1160.6 | Only Gemini 3.1 Pro Preview measured |
| LMArena Text: coding | 1454 | 1484 | Gemini 3.1 Pro Preview |
| LMArena Text: software and IT | 1439 | 1487 | Gemini 3.1 Pro Preview |
| LMArena WebDev: React | 1326 | 1438 | Gemini 3.1 Pro Preview |
| LMArena WebDev: HTML | 1318 | 1490 | Gemini 3.1 Pro Preview |
| LMArena WebDev: from an image | Not measured | 1481 | Only Gemini 3.1 Pro Preview measured |
| Neither model has a result on the other 12 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 16.9 | 29.7 | Gemini 3.1 Pro Preview |
| Epoch Capabilities Index | 142.2 | Not measured | Only Claude Haiku 4.5 measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index08
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| AA Agentic Index | 8.0 | 8.2 | Gemini 3.1 Pro Preview |
| Design Arena: agents building web apps | Not measured | 1125 | Only Gemini 3.1 Pro Preview measured |
| Design Arena: agents building full-stack apps | Not measured | 1068 | Only Gemini 3.1 Pro Preview measured |
| Design Arena: agents building mobile apps | Not measured | 1094 | Only Gemini 3.1 Pro Preview measured |
| Design Arena: agents building Android apps | Not measured | 1055 | Only Gemini 3.1 Pro Preview measured |
| Design Arena: agents building Godot games | Not measured | 1236 | Only Gemini 3.1 Pro Preview measured |
| Design Arena: agents building games | Not measured | 1114 | Only Gemini 3.1 Pro Preview measured |
| Design Arena: HTML slides | Not measured | 1153 | Only Gemini 3.1 Pro Preview measured |
| Design Arena: PowerPoint slides | Not measured | 1064 | Only Gemini 3.1 Pro Preview measured |
| LiveBench research | 42.2% | Not measured | Only Claude Haiku 4.5 measured |
| LiveBench browser | 40.0% | Not measured | Only Claude Haiku 4.5 measured |
| LiveBench long agentic | 11.1% | Not measured | Only Claude Haiku 4.5 measured |
| SREBench (Vals AI) | Not measured | 0.0% | Only Gemini 3.1 Pro Preview measured |
| Finance Agent (Vals AI) | 31.0% | 43.0% | Gemini 3.1 Pro Preview |
| Tax Agent Bench (Vals AI) | 5.9% | 9.3% | Gemini 3.1 Pro Preview |
| Legal Research Bench (Vals AI) | 10.6% | 20.7% | Gemini 3.1 Pro Preview |
| Excel Modeling Benchmark (Vals AI) | 22.7% | 52.6% | Gemini 3.1 Pro Preview |
| BioMysteryBench (Vals AI) | Not measured | 60.0% | Only Gemini 3.1 Pro Preview measured |
| Terminal-Bench-Science (Vals AI) | Not measured | 1.4% | Only Gemini 3.1 Pro Preview measured |
| Vals Science Index | Not measured | 27.4% | Only Gemini 3.1 Pro Preview measured |
| τ²-Bench Telecom | Not measured | 95.6% | Only Gemini 3.1 Pro Preview measured |
| APEX-Agents | Not measured | 35.1% | Only Gemini 3.1 Pro Preview measured |
| METR time horizons | Not measured | 77.0% | Only Gemini 3.1 Pro Preview measured |
| BALROG | 31.2% | 57.0% | Gemini 3.1 Pro Preview |
| DeepResearch Bench | 45.5% | 47.8% | Gemini 3.1 Pro Preview |
| PostTrainBench | Not measured | 22.0% | Only Gemini 3.1 Pro Preview measured |
| Vending-Bench 2 | $459 | $911 | Gemini 3.1 Pro Preview |
| Neither model has a result on the other 12 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index09
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| GPQA Diamond | 72.2% | 95.5% | Gemini 3.1 Pro Preview |
| LiveBench reasoning | 61.7% | Not measured | Only Claude Haiku 4.5 measured |
| MysteryMechanism (Vals AI) | Not measured | 20.7% | Only Gemini 3.1 Pro Preview measured |
| MMMU Pro (Vals AI) | 46.1% | 88.2% | Gemini 3.1 Pro Preview |
| GPQA Diamond (Artificial Analysis run) | Not measured | 94.1% | Only Gemini 3.1 Pro Preview measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 47.0% | Only Gemini 3.1 Pro Preview measured |
| GPQA Diamond (Epoch AI run) | 71.2% | 94.4% | Gemini 3.1 Pro Preview |
| ARC-AGI | 47.7% | 98.0% | Gemini 3.1 Pro Preview |
| ARC-AGI-2 | 4.0% | 77.1% | Gemini 3.1 Pro Preview |
| SimpleBench | Not measured | 79.6% | Only Gemini 3.1 Pro Preview measured |
| Chess puzzles | 8.0% | 51.0% | Gemini 3.1 Pro Preview |
| Mystery game puzzles | Not measured | 34.0% | Only Gemini 3.1 Pro Preview measured |
| EBR-bench | Not measured | 14.3% | Only Gemini 3.1 Pro Preview measured |
| LMArena Text: hard prompts | 1421 | 1485 | Gemini 3.1 Pro Preview |
| LMArena Text: expert | 1443 | 1483 | Gemini 3.1 Pro Preview |
| LMArena Text: hard prompts in English | 1435 | 1480 | Gemini 3.1 Pro Preview |
| LMArena Vision: diagrams | Not measured | 1311 | Only Gemini 3.1 Pro Preview measured |
| LMArena Vision: homework | Not measured | 1316 | Only Gemini 3.1 Pro Preview measured |
| Neither model has a result on the other 3 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index04
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| AIME | 82.7% | 98.1% | Gemini 3.1 Pro Preview |
| MGSM (Vals AI) | 92.2% | Not measured | Only Claude Haiku 4.5 measured |
| OTIS Mock AIME | 66.7% | 95.6% | Gemini 3.1 Pro Preview |
| FrontierMath, tiers 1 to 3 | Not measured | 59.6% | Only Gemini 3.1 Pro Preview measured |
| FrontierMath, tier 4 | Not measured | 26.8% | Only Gemini 3.1 Pro Preview measured |
| ProofBench | Not measured | 26.0% | Only Gemini 3.1 Pro Preview measured |
| MATH level 5 | 96.4% | Not measured | Only Claude Haiku 4.5 measured |
| LMArena Text: math | 1398 | 1486 | Gemini 3.1 Pro Preview |
| LMArena Text: mathematical work | 1430 | 1478 | Gemini 3.1 Pro Preview |
| Neither model has a result on the other 4 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index013
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| MMLU Pro | 78.7% | 91.0% | Gemini 3.1 Pro Preview |
| LegalBench (Vals AI) | 81.2% | 87.4% | Gemini 3.1 Pro Preview |
| CaseLaw (Vals AI) | 56.5% | 64.8% | Gemini 3.1 Pro Preview |
| CorpFin (Vals AI) | 60.6% | 64.5% | Gemini 3.1 Pro Preview |
| TaxEval (Vals AI) | 67.5% | 72.9% | Gemini 3.1 Pro Preview |
| MortgageTax (Vals AI) | 62.2% | 69.4% | Gemini 3.1 Pro Preview |
| MedQA (Vals AI) | 79.6% | 96.4% | Gemini 3.1 Pro Preview |
| MedCode (Vals AI) | 32.7% | 59.1% | Gemini 3.1 Pro Preview |
| SimpleQA Verified | 13.2% | 73.5% | Gemini 3.1 Pro Preview |
| LMArena Text: science | 1406 | 1497 | Gemini 3.1 Pro Preview |
| LMArena Text: medicine and healthcare | 1394 | 1486 | Gemini 3.1 Pro Preview |
| LMArena Text: legal and government | 1400 | 1487 | Gemini 3.1 Pro Preview |
| LMArena Text: business and finance | 1400 | 1459 | Gemini 3.1 Pro Preview |
| Neither model has a result on the other 4 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index08
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| Design Arena: websites | 1122 | 1234 | Gemini 3.1 Pro Preview |
| Design Arena: UI components | 1100 | 1226 | Gemini 3.1 Pro Preview |
| Design Arena: game dev | 1097 | 1203 | Gemini 3.1 Pro Preview |
| Design Arena: data viz | 1124 | 1209 | Gemini 3.1 Pro Preview |
| Design Arena: 3D | 1072 | 1234 | Gemini 3.1 Pro Preview |
| Design Arena: all code categories | 1115 | 1231 | Gemini 3.1 Pro Preview |
| Design Arena: SVG | 1025 | 1285 | Gemini 3.1 Pro Preview |
| Design Arena: ASCII art | 1139 | 1270 | Gemini 3.1 Pro Preview |
| LiveBench design | 59.9% | Not measured | Only Claude Haiku 4.5 measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index223
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| LiveBench data analysis | 59.3% | Not measured | Only Claude Haiku 4.5 measured |
| Public Benefits Bench (Vals AI) | 54.3% | 53.8% | Claude Haiku 4.5 |
| SAGE (Vals AI) | 31.8% | 48.7% | Gemini 3.1 Pro Preview |
| MedScribe (Vals AI) | 85.2% | 76.1% | Claude Haiku 4.5 |
| Vals Multimodal Index | 42.9% | 56.1% | Gemini 3.1 Pro Preview |
| IFBench | Not measured | 77.1% | Only Gemini 3.1 Pro Preview measured |
| Long context reasoning (AA-LCR) | Not measured | 82.0% | Only Gemini 3.1 Pro Preview measured |
| DTBench | 73.6% | 97.1% | Gemini 3.1 Pro Preview |
| LMCA | 30.9% | 53.8% | Gemini 3.1 Pro Preview |
| CL-bench | Not measured | 20.8% | Only Gemini 3.1 Pro Preview measured |
| Furniture assembly | Not measured | 26.7% | Only Gemini 3.1 Pro Preview measured |
| ExploitBench | 13.7% | 26.1% | Gemini 3.1 Pro Preview |
| CL-bench Life | Not measured | 16.9% | Only Gemini 3.1 Pro Preview measured |
| LMArena Text | 1397 | 1480 | Gemini 3.1 Pro Preview |
| LMArena Text: instruction following | 1414 | 1466 | Gemini 3.1 Pro Preview |
| LMArena Text: creative writing | 1371 | 1482 | Gemini 3.1 Pro Preview |
| LMArena Vision | Not measured | 1296 | Only Gemini 3.1 Pro Preview measured |
| LMArena Text: English | 1414 | 1478 | Gemini 3.1 Pro Preview |
| LMArena Text: non-English | 1377 | 1476 | Gemini 3.1 Pro Preview |
| LMArena Text: Chinese | 1418 | 1527 | Gemini 3.1 Pro Preview |
| LMArena Text: German | 1374 | 1490 | Gemini 3.1 Pro Preview |
| LMArena Text: Japanese | 1338 | 1491 | Gemini 3.1 Pro Preview |
| LMArena Text: Korean | 1346 | 1455 | Gemini 3.1 Pro Preview |
| LMArena Text: French | 1411 | 1487 | Gemini 3.1 Pro Preview |
| LMArena Text: Spanish | 1421 | 1481 | Gemini 3.1 Pro Preview |
| LMArena Text: Russian | 1381 | 1497 | Gemini 3.1 Pro Preview |
| LMArena Text: Polish | 1343 | 1498 | Gemini 3.1 Pro Preview |
| LMArena Text: multi-turn | 1409 | 1487 | Gemini 3.1 Pro Preview |
| LMArena Text: longer queries | 1427 | 1482 | Gemini 3.1 Pro Preview |
| LMArena Text: writing and language | 1386 | 1481 | Gemini 3.1 Pro Preview |
| LMArena Text: entertainment and media | 1369 | 1465 | Gemini 3.1 Pro Preview |
| LMArena Vision: English | Not measured | 1287 | Only Gemini 3.1 Pro Preview measured |
| LMArena Vision: Chinese | Not measured | 1364 | Only Gemini 3.1 Pro Preview measured |
| LMArena Vision: reading text | Not measured | 1304 | Only Gemini 3.1 Pro Preview measured |
| LMArena Vision: humor | Not measured | 1322 | Only Gemini 3.1 Pro Preview measured |
| LMArena Vision: captioning | Not measured | 1345 | Only Gemini 3.1 Pro Preview measured |
| LMArena Vision: recognizing things | Not measured | 1330 | Only Gemini 3.1 Pro Preview measured |
| LMArena Vision: creative writing | Not measured | 1324 | Only Gemini 3.1 Pro Preview measured |
| LMArena Document | 1420 | 1444 | Gemini 3.1 Pro Preview |
| Neither model has a result on the other 7 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Claude Haiku 4.5 | Gemini 3.1 Pro Preview | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Claude Haiku 4.5 costs less
Repository review
50K input + 3K output tokens
Claude Haiku 4.5 costs less
Agent loop
200K input + 10K output tokens
Claude Haiku 4.5 costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Anthropic Claude Haiku 4.5 | Google Gemini 3.1 Pro Preview |
|---|---|---|
| Maker | Anthropic | |
| Weights | No public weights listed | No public weights listed |
| Released | October 15, 2025 | February 19, 2026 |
| Context window | 200K | 1.05M |
| Input $/Mtok | $1.00 | $2.00 |
| Output $/Mtok | $5.00 | $12.00 |
| Coding index | 49.8 | 35.3 |
| Independent measurements behind it | 3 | 1 |
| Cost per solved task | $0.56 | $0.99 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Claude Haiku 4.5 better than Gemini 3.1 Pro Preview for coding?
Pick Claude Haiku 4.5 for coding: it scores 49.8 on the Vibe Coding Index against 35.3 for Gemini 3.1 Pro Preview. Pick Gemini 3.1 Pro Preview when the context window is larger, 1.05M against 200K tokens, or when what matters is LMArena WebDev or AA Coding Index, where it leads.
Which is cheaper, Claude Haiku 4.5 or Gemini 3.1 Pro Preview?
Claude Haiku 4.5 costs $1.00 per million input tokens and $5.00 per million output tokens; Gemini 3.1 Pro Preview costs $2.00 and $12.00. Per solved SWE-bench Verified task, Claude Haiku 4.5 costs $0.56 and Gemini 3.1 Pro Preview $0.99.
Which has the bigger context window, Claude Haiku 4.5 or Gemini 3.1 Pro Preview?
Gemini 3.1 Pro Preview, with 1.05M tokens against 200K.
How many benchmarks is the Claude Haiku 4.5 vs Gemini 3.1 Pro Preview comparison based on?
82 benchmarks have a result for both models as of October 21, 2026. Claude Haiku 4.5's coding index rests on 3 independent benchmarks and Gemini 3.1 Pro Preview's on 1. Where only one model has a result, the page says so instead of naming a winner.