GPT-5.4 Pro vs Kimi K3
Full profiles: GPT-5.4 Pro (OpenAI) and Kimi K3 (Moonshot AI): every benchmark, price and rank.
Updated · Data from 9 sources
As of 25 Oct 2026, GPT-5.4 Pro and Kimi K3 (max) are neck and neck on quality (63.3 vs 64.8 on the BenchLeader Index, inside the ±4.7 margin of error), and Kimi K3 (max) costs 11× less, which makes it the better buy for most workloads. GPT-5.4 Pro leads in every category we measure both in.
GPT-5.4 Pro benchmarks 
Kimi K3 benchmarks Which to pick for…
The pick's numbers come first. Category scores on the BenchLeader scale (50 is the average model, 15 points one standard deviation). High confidence means a lead of half a standard deviation or more; under one point is called either.
| Use case | Pick | Why | Confidence |
|---|---|---|---|
| Reasoning | GPT-5.4 Pro | 73 vs 66 category score; SimpleBench 74.1% vs 60.7% | Medium |
| Knowledge | GPT-5.4 Pro | 74 vs 63 category score | High |
| Images and other media | GPT-5.4 Pro | 70 vs 64 category score | Medium |
| Chat quality (human preference) | – | Only Kimi K3 has an Arena rating (68 category score) | Not comparable |
| Lowest cost | Kimi K3 | $6.00 vs $67.50 per 1M blended; agentic coding session $0.240 vs $2.52 | – |
| Fastest first answer | GPT-5.4 Pro | 6.1 s (GPT-5.4 Pro) vs 52.6 s (Kimi K3 (max)) to the first answer token | – |
| High volume on a budget | Neither flagship | A smaller model is usually the better fit: Cheapest good LLMs, Fastest LLMs | – |
Key differences
- GPT-5.4 Pro and Kimi K3 (max) are neck and neck on quality (63.3 vs 64.8, within the ±4.7 margin).
- GPT-5.4 Pro is stronger in instruction following, knowledge, multimodal, reasoning.
- Kimi K3 (max) is stronger in agents & tools, coding, composite, human preference, long context, maths.
- Kimi K3 (max) is 11× cheaper ($6.00 vs $67.50 per 1M blended).
- Kimi K3 (max) streams 5.8× faster (41 vs 7 tokens per second).
At a glance
| Spec | GPT-5.4 Pro | Kimi K3 |
|---|---|---|
| Developer | OpenAI | Moonshot AI |
| Released | 4 Mar 2026 | 16 Jul 2026 |
| Context window (tokens) | 1,050,000 | 1,048,576 |
| Compared at | GPT-5.4 Pro | Kimi K3 (max) |
| Open weights | No | Yes |
| Price in / out / cached input per 1M | $30.00 / $180.00 / – | $3.00 / $15.00 / $0.300 |
Category by category
Category scores on the BenchLeader scale: 50 is the average evaluated model, 15 points is one standard deviation. A gap under one point is called level.
| Category | GPT-5.4 Pro | Kimi K3 (max) | Leader |
|---|---|---|---|
| Reasoning | 72.5 | 66.4 | GPT-5.4 Pro |
| Coding | – | 64.6 | not comparable |
| Agents & tools | – | 59.4 | not comparable |
| Maths | – | 63.4 | not comparable |
| Knowledge | 73.6 | 62.8 | GPT-5.4 Pro |
| Instruction following | 63.8 | – | not comparable |
| Human preference | – | 68.1 | not comparable |
| Multimodal | 69.8 | 63.6 | GPT-5.4 Pro |
| Long context | – | 69.6 | not comparable |
| Composite | – | 78.3 | not comparable |
Where sources disagree
Across the 2 benchmarks both have, GPT-5.4 Pro leads 1, Kimi K3 leads 1.
| Overall index | GPT-5.4 Pro | Kimi K3 (max) | Leader |
|---|---|---|---|
| BenchLeader Index | 63.3 | 64.8 | neck and neck |
Each publisher runs its own tasks, harnesses and effort settings, so their overall verdicts can differ; BenchLeader reconciles them on one scale. † measured at another effort setting of the same model (the publisher did not test the one compared here).
Price, speed and latency
List prices blended at three input tokens per output token. Per-task prices use the token counts on the methodology page; suite cost is what Artificial Analysis paid to run its whole Intelligence Index.
| Measure | GPT-5.4 Pro | Kimi K3 (max) |
|---|---|---|
| Blended price per 1M tokens | $67.50 | $6.00 |
| Input / output per 1M | $30.00 / $180.00 | $3.00 / $15.00 |
| Cached input per 1M | – | $0.300 |
| Price per task: chat reply | $0.066 | $0.0057 |
| Price per task: summarize a 30-page report | $0.468 | $0.045 |
| Price per task: code edit | $0.450 | $0.041 |
| Price per task: agentic coding session | $2.52 | $0.240 |
| Price per task: structured extraction | $0.096 | $0.0090 |
| Cost to run a full benchmark suite | – | $2.00 |
| Output speed | 7 tok/s | 41 tok/s |
| Time to first answer | 6.1 s | 52.6 s |
| Time to a complete answer | – | 65.0 s |
Each model is compared at its best-ranked reasoning-effort setting. GPT-5.4 Pro scores 63.3 and takes 6 s to its first answer token; GPT-5.4 Pro (xhigh) scores 61.9 in 6 s. Kimi K3 (max) scores 64.8 and takes 53 s to its first answer token; Kimi K3 scores 58.8 in 1 s, Kimi K3 (low) scores 58.2 in 57 s. Higher effort means more thinking before the answer: better scores, longer waits and more tokens billed. How reasoning effort changes scores and cost.
Every benchmark
137 benchmarks, grouped by category; each name links to the board and its leaders. † measured at another effort setting of the same model, shown for reference and not counted as a win.
| Metric | Publisher | GPT-5.4 Pro | Kimi K3 (max) |
|---|---|---|---|
| BenchLeader Index | BenchLeader | 63.3 | 64.8 |
| Instruction following score | BenchLeader | 63.8 | – |
| Knowledge score | BenchLeader | 73.6 | 62.8 |
| Multimodal score | BenchLeader | 69.8 | 63.6 |
| Reasoning score | BenchLeader | 72.5 | 66.4 |
| Agents & tools score | BenchLeader | – | 59.4 |
| Coding score | BenchLeader | – | 64.6 |
| Composite score | BenchLeader | – | 78.3 |
| Human preference score | BenchLeader | – | 68.1 |
| Long context score | BenchLeader | – | 69.6 |
| Maths score | BenchLeader | – | 63.4 |
| Blended price $/M | BenchLeader | $67.50 | $6.00 |
| Output speed | BenchLeader | 7 tok/s | 41 tok/s |
| Time to first answer | BenchLeader | 6.1 s | 52.6 s |
| Context window | BenchLeader | 1,050,000 tokens | 1,048,576 tokens |
| Reasoning | |||
| GPQA Diamond | Epoch AI Benchmarking Hub | 94.6%† | 93.1% |
| Humanity's Last Exam | Scale AI / CAIS | 44.3% | – |
| SimpleBench | SimpleBench | 74.1% | 60.7% |
| Arena Hard Prompts | Arena (LMArena) | – | 1516 |
| LiveBench 2.0 Reasoning | LiveBench | – | 90.7%† |
| LiveBench 1.0 Reasoning | LiveBench | – | 90.7%† |
| GPQA Diamond (AA) | Artificial Analysis | – | 93.5% |
| Humanity's Last Exam (AA) | Artificial Analysis | – | 46.9% |
| GPQA Diamond (Vals) | Vals AI | – | 92.9%† |
| ARC-AGI-1 | ARC Prize | 94.5%† | 94.5% |
| ARC-AGI-2 | ARC Prize | 83.3%† | 60.4% |
| CritPt | Artificial Analysis | 30.0%† | 23.4% |
| Chess Puzzles | Epoch AI Benchmarking Hub | 59.0%† | 36.0% |
| Mystery Game Puzzles | Epoch AI Benchmarking Hub | – | 26.0% |
| LMCA | Epoch AI Benchmarking Hub | – | 52.7% |
| DTBench | Epoch AI Benchmarking Hub | – | 91.2% |
| ForecastBench | ForecastBench | – | 64.6% |
| Coding | |||
| SciCode | SciCode | – | 59.5% |
| WeirdML | WeirdML | 57.4%† | 82.6% |
| FrontierCode | Cognition | – | 44.2%† |
| Arena Coding | Arena (LMArena) | – | 1539 |
| Arena WebDev | Arena (LMArena) | – | 1655 |
| LiveBench 2.0 Coding | LiveBench | – | 59.3%† |
| LiveBench 1.0 Coding | LiveBench | – | 81.5%† |
| LiveCodeBench | Vals AI | – | 87.2% |
Show all 137 benchmarks
| Metric | Publisher | GPT-5.4 Pro | Kimi K3 (max) |
|---|---|---|---|
| Coding | |||
| IOI | Vals AI | – | 48.9% |
| SWE-bench (Vals) | Vals AI | – | 93.4%† |
| Code Migration | Vals AI | – | 16.1% |
| Code Migration v1.1 | Vals AI | – | 28.6% |
| ProgramBench | Vals AI | – | 2.0%† |
| SWE-bench Multimodal | Vals AI | – | 44.2% |
| Vibe Code Bench 1-100 | Vals AI | – | 18.2% |
| Vibe Code Bench v1.1 | Vals AI | – | 85.0% |
| Surface Evolver Bench | Epoch AI Benchmarking Hub | – | 93.0% |
| DeepSWE v1.1 | Epoch AI Benchmarking Hub | – | 68.5% |
| ALE-Bench | Epoch AI Benchmarking Hub | – | 1524.5 |
| FrontierSWE | Epoch AI Benchmarking Hub | – | 25.9% |
| Terminal-Bench 4.0 (AA) | Artificial Analysis | – | 12.6% |
| Terminal-Bench 2.1 (AA) | Artificial Analysis | – | 85.0% |
| Startrise Frontend | Startrise | – | 79.8%† |
| FrontierCode 1.1 (AA) | Artificial Analysis | – | 43.6% |
| Agents & tools | |||
| APEX-Agents | Mercor | – | 47.8%† |
| Arena Agent | Arena (LMArena) | – | 1.6 |
| LiveBench Long Agentic | LiveBench | – | 42.8%† |
| LiveBench Browser | LiveBench | – | 16.0%† |
| LiveBench Agentic Coding | LiveBench | – | 62.2%† |
| Terminal-Bench 2.1 (Vals) | Vals AI | – | 80.9%† |
| MCP Atlas | Scale AI SEAL | – | 82.3% |
| HiL-Bench | Scale AI SEAL | – | 42.7%† |
| τ³-bench Banking | τ²-bench | – | 37.1% |
| τ³-Banking (AA) | Artificial Analysis | – | 46.0% |
| ITBench SRE (AA) | Artificial Analysis | – | 47.7% |
| Analyst Agent (AA) | Artificial Analysis | – | 38.8% |
| APEX-Agents (AA) | Artificial Analysis | – | 41.3% |
| CyberBench | Vals AI | – | 75.2% |
| Finance Agent v2 | Vals AI | – | 53.1% |
| Finance Agent v2.1 | Vals AI | – | 60.0% |
| Harvey's Legal Agent Benchmark v1.1 | Vals AI | – | 12.9% |
| Harvey's Legal Agent Benchmark v1 | Vals AI | – | 12.9% |
| Legal Research Bench | Vals AI | – | 46.1% |
| Tax Agent Bench | Vals AI | – | 33.4% |
| Terminal-Bench 4.0 (Vals) | Vals AI | – | 17.2% |
| Terminal-Bench Science | Vals AI | – | 1.4% |
| Time Horizon Index: KSP | Vals AI | – | 10.5% |
| PostTrainBench | Epoch AI Benchmarking Hub | – | 32.0%† |
| GBAEval | Epoch AI Benchmarking Hub | – | 48.4%† |
| Vending-Bench 2 | Andon Labs | – | 5165.0† |
| AutomationBench | Artificial Analysis | – | 58.3% |
| GDP.pdf (AA) | Artificial Analysis | – | 22.0% |
| Harvey LAB-AA v1.1 | Harvey | – | 5.3% |
| EnterpriseOps-Gym | Artificial Analysis | – | 45.3% |
| AA-Briefcase v1.1 | Artificial Analysis | – | 1499 |
| Terminal-Bench-Science 0.1 (AA) | Artificial Analysis | – | 8.1% |
| CWE-Bench (AA) | Artificial Analysis | – | 53.3% |
| DeepSec-Bench (AA) | Artificial Analysis | – | 12.4% |
| CyberGym end-to-end (AA) | Artificial Analysis | – | 58.0% |
| Maths | |||
| FrontierMath Tiers 1–3 | Epoch AI Benchmarking Hub | 82.5%† | 72.2% |
| FrontierMath Tier 4 | Epoch AI Benchmarking Hub | 58.5%† | 39.0% |
| OTIS Mock AIME | Epoch AI Benchmarking Hub | – | 97.2% |
| ProofBench | Vals AI | – | 87.0%† |
| LiveBench Mathematics | LiveBench | – | 84.4%† |
| AIME 2026 | MathArena | – | 96.7%† |
| HMMT February 2026 | MathArena | – | 97.0%† |
| MathArena Apex | MathArena | 69.8%† | 65.6%† |
| Arena Maths | Arena (LMArena) | – | 1499 |
| Knowledge | |||
| SimpleQA Verified | Epoch AI Benchmarking Hub | 46.3%† | 50.6% |
| LiveBench 2.0 Data Analysis | LiveBench | – | 78.7%† |
| LiveBench Research | LiveBench | – | 79.0%† |
| LiveBench 1.0 Data Analysis | LiveBench | – | 78.7%† |
| AA-Omniscience | Artificial Analysis | – | 19.7 |
| MMLU-Pro | Vals AI | – | 88.0% |
| LegalBench | Vals AI | – | 86.2% |
| CorpFin | Vals AI | – | 71.6%† |
| TaxEval | Vals AI | – | 75.7%† |
| MultiNRC | Scale AI SEAL | 62.3% | – |
| BioMysteryBench | Vals AI | – | 72.6% |
| Excel Modeling Benchmark | Vals AI | – | 66.7% |
| Excel Modeling Benchmark v1.1 | Vals AI | – | 71.3% |
| MedCode | Vals AI | – | 49.4% |
| MedScribe | Vals AI | – | 88.0% |
| Public Benefits Bench v1.1 | Vals AI | – | 68.2% |
| GDP.pdf | Epoch AI Benchmarking Hub | – | 19.0% |
| AA-Omniscience: accuracy | Artificial Analysis | – | 47.6% |
| AA-Omniscience: non-hallucination | Artificial Analysis | – | 46.8% |
| Instruction following | |||
| LiveBench Language | LiveBench | – | 85.5%† |
| LiveBench Instruction Following | LiveBench | – | 71.4%† |
| MultiChallenge | Scale AI SEAL | 69.2% | – |
| TutorBench | Scale AI SEAL | 56.6% | – |
| Arena Instruction Following | Arena (LMArena) | – | 1489 |
| Human preference | |||
| Arena Text | Arena (LMArena) | – | 1487 |
| EQ-Bench 4 | EQ-Bench | – | 1318† |
| Arena Creative Writing | Arena (LMArena) | – | 1462 |
| Arena Multi-turn | Arena (LMArena) | – | 1502 |
| Arena Longer Queries | Arena (LMArena) | – | 1503 |
| Multimodal | |||
| LiveBench Design | LiveBench | – | 77.2%† |
| MMMU-Pro | Artificial Analysis | – | 80.5% |
| VISTA | Scale AI SEAL | 53.9% | – |
| MMMU-Pro (Vals) | Vals AI | – | 88.2%† |
| MortgageTax | Vals AI | – | 66.3%† |
| SAGE | Vals AI | – | 52.8% |
| Blueprint-Bench 2 | Epoch AI Benchmarking Hub | – | 29.5%† |
| AA-3D | Artificial Analysis | – | 982 |
| Long context | |||
| AA-LCR | Artificial Analysis | – | 88.7% |
| MLCR | Artificial Analysis | – | 21.7% |
| Composite | |||
| Epoch Capabilities Index | Epoch AI Benchmarking Hub | 159.6 | 157.9† |
| LiveBench 2.0 | LiveBench | – | 63.4%† |
| LiveBench 1.0 | LiveBench | – | 79.2%† |
| AA Intelligence Index v4.3.2 | Artificial Analysis | – | 43.6 |
| Vals Index | Vals AI | – | 53.0 |
| Vals Finance Index | Vals AI | – | 58.0% |
| Vals RSI Index | Vals AI | – | 20.9% |
| Vals Multimodal Index | Vals AI | – | 73.4%† |
| Vals Router Index | Vals AI | – | 53.0%† |
| AA Cyber Index | Artificial Analysis | – | 41.3 |
| Safety & honesty | |||
| FORTRESS | Scale AI SEAL | 14.8% | 26.6% |
| MASK | Scale AI SEAL | 91.7% | – |
| Metadata | |||
| AA Openness Index | Artificial Analysis | – | 38.9 |
Data as of 2026-10-25. Best configuration of each model; every score links to its source on the model pages.
What changed
- Kimi K3 (max): 7 new results (Code Migration v1.1, Excel Modeling Benchmark v1.1, Finance Agent v2.1…).
- GPT-5.4 Pro: 1 score revision.
- Kimi K3 (max): 1 new result (Terminal-Bench-Science 0.1 (AA)); 1 score revision.
- Kimi K3 (max): 3 score revisions.
- Kimi K3 (max): 1 score revision.
- Kimi K3 (max): 1 new result (τ³-bench Banking).
- Kimi K3 (max): 8 new results (LegalBench, Vals Index, Finance Agent v2…); 5 score revisions.
- Kimi K3 (max): 3 score revisions.
- Kimi K3 (max): 1 score revision.
- Kimi K3 (max): BenchLeader Index 65.7 → 64.9.
Related comparisons
- GPT-5.4 Pro vs Claude Opus 5.5
- GPT-5.4 Pro vs Gemini 4 Argon
- GPT-5.4 Pro vs Muse Wonder Full
- Kimi K3 vs Claude Opus 5.5
- Kimi K3 vs Gemini 4 Argon
- Kimi K3 vs Muse Wonder Full
- Kimi K3 vs Claude Fable 5.1
- GPT-5.3-Codex vs Kimi K3
Rankings: Best LLM overall · Best LLM for agents · Best cheap LLMs
GPT-5.4 Pro vs Kimi K3: questions
- Is GPT-5.4 Pro better than Kimi K3?
- GPT-5.4 Pro and Kimi K3 (max) are neck and neck on quality (63.3 vs 64.8, within the ±4.7 margin). The BenchLeader Index combines every independent quality benchmark; the two are within the margin of error as of 2026-10-25, so check the category scores for your use.
- Which is cheaper, GPT-5.4 Pro or Kimi K3, and will that change?
- Kimi K3 is cheaper today: $6.00 against $67.50 per million tokens, blended at three input tokens per output token. No price change has been announced for either.
- Which is better for long documents?
- Both take about 1.1M tokens: GPT-5.4 Pro 1,050,000, Kimi K3 1,048,576.
- Which has lower latency?
- GPT-5.4 Pro answers first: 6.1 s against 52.6 s to the first answer token, as Artificial Analysis measures it. Reasoning effort drives most of this: a higher setting thinks longer before it answers.
- Which is faster, GPT-5.4 Pro or Kimi K3?
- Kimi K3 streams faster: 41 against 7 output tokens per second.
- Do independent benchmarks agree on GPT-5.4 Pro vs Kimi K3?
- Across the 2 benchmarks both have, GPT-5.4 Pro leads 1, Kimi K3 leads 1.
- When was Kimi K3 released?
- Kimi K3 was released on 16 Jul 2026.