Model comparison
Claude Haiku 4.5 vs GPT-5.4
GPT-5.4 is the stronger model overall, scoring 59.4 to 39.5 on the Noometry Index. Claude Haiku 4.5 costs 2.8× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Last verified . 42 shared benchmarks.
Summary
- They share 42 benchmarks with published results for both. Claude Haiku 4.5 scores higher in 0 categories and GPT-5.4 in 10 categories; 10 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where GPT-5.4 leads 61.8 to 15.1.
- The biggest single-benchmark swing is NYT Connections (extended): 14.3% for Claude Haiku 4.5 and 91.3% for GPT-5.4.
- Claude Haiku 4.5 is cheaper at $1 / $5 per million input/output tokens, against $2.50 / $15 for GPT-5.4.
- GPT-5.4 accepts more context: 1.05M tokens versus 200K.
Side by side
| Claude Haiku 4.5 | GPT-5.4 | |
|---|---|---|
| Provider | Anthropic | OpenAI |
| Noometry Index | 39.5 | 59.4 |
| Released | 2025-10-15 | 2026-03-05 |
| Weights | Proprietary | Proprietary |
| Context window | 200K | 1.05M |
| Max output | 64K | 128K |
| Input $ / M tokens | $1 | $2.50 |
| Output $ / M tokens | $5 | $15 |
| Results tracked | 53 | 68 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding GPT-5.4 leads
Claude Haiku 4.5: 44.0 (#78), GPT-5.4: 52.6 (#33)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| LMArena WebDev | 1330 | 1465 |
| SciCode | 43.3% | 56.6% |
| WeirdML | 45.4% | 77.7% |
| LMArena Coding | 1453 | 1497 |
| ALE-Bench | 653.48 | 1,607 |
| SWE-bench Verified | — | 76.9% |
| DeepSWE | — | 51.8% |
| SWE-bench Verified (bash only) | 66.6% | — |
| SWE-bench Multilingual | 64.7% | — |
| GSO | — | 31.4% |
| MirrorCode | — | 15.6% |
| AlgoTune | — | 1.85 |
Agentic & Tool Use GPT-5.4 leads
Claude Haiku 4.5: 33.6 (#52), GPT-5.4: 46.5 (#13)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| Terminal-Bench | 35.5% | 81.8% |
| DeepResearch Bench | 45.5% | 35.1% |
| Vending-Bench 2 | 458.89 | 6,144 |
| APEX-Agents | — | 52.4% |
| Berkeley Function Calling Leaderboard | 68.7% | — |
| τ²-bench Banking | — | 39.4% |
| PostTrainBench | — | 19% |
| BALROG | 31.2% | — |
| ExploitBench | 13.7% | — |
| GBAEval | — | 45.1% |
| LMArena Search | — | 1197 |
| METR Time Horizons | — | 74.3% |
Reasoning GPT-5.4 leads
Claude Haiku 4.5: 15.1 (#320), GPT-5.4: 61.8 (#19)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| ARC-AGI-2 | 4% | 74% |
| NYT Connections (extended) | 14.3% | 91.3% |
| ARC-AGI-1 | 47.7% | 93.7% |
| CritPt | 0% | 23.4% |
| Chess Puzzles | 8% | 44% |
| LMArena Hard Prompts | 1420 | 1485 |
| DTBench | 73.6% | 94.4% |
| LMCA | 30.9% | 52% |
| Epoch Capabilities Index | 142.41 | 156.81 |
| ForecastBench | 61.4 | 59.5 |
| Kagi LLM Benchmark | — | 63.8% |
| EnigmaEval | — | 16% |
| Thematic Generalization | — | 80% |
| EBR-Bench | — | 25.4% |
| Mystery Game Puzzles | — | 37% |
Math GPT-5.4 leads
Claude Haiku 4.5: 44.9 (#78), GPT-5.4: 73.5 (#19)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 66.7% | 97.8% |
| LMArena Math | 1396 | 1488 |
| FrontierMath (Feb 2025 set) | 5.9% | 47.6% |
| FrontierMath Tier 4 (v1) | 2.1% | 27.1% |
| FrontierMath (Tiers 1-3) | — | 78.6% |
| FrontierMath Tier 4 | — | 49% |
| MathArena Final-Answer Competitions | — | 83.1% |
| ProofBench | — | 56% |
| Omni-MATH | 56.1% | — |
| MATH Level 5 | 96.4% | — |
Knowledge GPT-5.4 leads
Claude Haiku 4.5: 37.7 (#153), GPT-5.4: 65.3 (#14)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| GPQA Diamond | 71.2% | 93.3% |
| SimpleQA Verified | 13.2% | 45.1% |
| Vectara Hallucination Rate | 9.8% | 7% |
| LMArena Expert | 1442 | 1507 |
| Humanity's Last Exam | — | 36.2% |
| MMLU-Pro | 77.7% | — |
| GPQA (HELM) | 60.5% | — |
Multimodal GPT-5.4 leads
Claude Haiku 4.5: 26.8 (#118), GPT-5.4: 43.7 (#20)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| Blueprint-Bench 2 | 0% | 27.1% |
| LMArena Document | 1420 | 1471 |
| LMArena Vision | — | 1303 |
| Furniture Assembly | — | 37.5% |
Multilingual GPT-5.4 leads
Claude Haiku 4.5: 49.9 (#129), GPT-5.4: 56.2 (#23)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| LMArena Non-English | 1377 | 1465 |
| LMArena Chinese | 1417 | 1519 |
| LMArena French | 1408 | 1493 |
| LMArena German | 1375 | 1472 |
| LMArena Japanese | 1339 | 1485 |
| LMArena Korean | 1347 | 1448 |
| LMArena Russian | 1381 | 1480 |
| LMArena Spanish | 1420 | 1454 |
Instruction Following GPT-5.4 leads
Claude Haiku 4.5: 71.4 (#149), GPT-5.4: 77.1 (#27)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| LMArena Instruction Following | 1414 | 1469 |
| IFEval | 80.1% | — |
Long Context GPT-5.4 leads
Claude Haiku 4.5: 43.6 (#92), GPT-5.4: 50.3 (#8)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| LMArena Longer Query | 1427 | 1473 |
| CL-bench | — | 27.9% |
| CL-bench Life | — | 21.7% |
Writing & Preference GPT-5.4 leads
Claude Haiku 4.5: 57.9 (#123), GPT-5.4: 71.9 (#17)
| Benchmark | Claude Haiku 4.5 | GPT-5.4 |
|---|---|---|
| LMArena Text | 1396 | 1469 |
| LMArena Creative Writing | 1372 | 1439 |
| EQ-Bench 4 | 1064 | 1272 |
| LMArena Multi-Turn | 1409 | 1482 |
| EQ-Bench Creative Writing | — | 1840 |
| WildBench | 83.9% | — |
Frequently asked questions
Is Claude Haiku 4.5 better than GPT-5.4?
GPT-5.4 is the stronger model overall, scoring 59.4 to 39.5 on the Noometry Index. Claude Haiku 4.5 costs 2.8× less per token, which makes it the better buy when GPT-5.4's lead doesn't matter for your workload.
Which is cheaper, Claude Haiku 4.5 or GPT-5.4?
Claude Haiku 4.5 is cheaper. It lists at $1 per million input tokens and $5 per million output tokens; GPT-5.4 lists at $2.50 and $15.
Is Claude Haiku 4.5 or GPT-5.4 better for coding?
GPT-5.4 scores higher on coding benchmarks: 52.6 versus 44.0 in the Noometry coding category.
Which has the bigger context window?
GPT-5.4 does, with 1.05M tokens against 200K.
How many benchmarks do Claude Haiku 4.5 and GPT-5.4 share?
42 benchmarks have published results for both models. Claude Haiku 4.5 has 53 scored results on Noometry and GPT-5.4 has 68.