Compare / head-to-head
GPT-4ovsKimi K3
Kimi K3 leads 4 of 4 shared benchmarks. GPT-4o is 1.4x cheaper per token. Kimi K3 has the larger context window (1M tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
0 – 4
Kimi K3 leads
Cheaper per token
GPT-4o
1.4x cheaper, input + output
Larger context
Kimi K3
1M tokens
Provider uptime (30d)
94.2446% · 100%
OpenAI · Moonshot AI
GPT-4o
OpenAI · released 2024-05-13
textvision
Quality
Benchmark matrix
4 shared · 4 only GPT-4o · 53 only Kimi K3| Benchmark | GPT-4o | Kimi K3 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 4 | ||||
| GPQA Diamond | 46% ↗ | 93.5% ↗ | -47.5 pt | Kimi K3 |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 0.4% ↗epoch | 72.2% ↗epoch | -71.8 pt | Kimi K3 |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 6.4% ↗epoch | 97.2% ↗epoch | -90.8 pt | Kimi K3 |
| SimpleQA Verified | 26% ↗epoch | 50.6% ↗epoch | -24.6 pt | Kimi K3 |
| Only GPT-4o reports · 4 | ||||
| MMLU | 85.7% ↗ | not reported | — | — |
| AIME 2025 | 11.7% ↗matharena | not reported | — | — |
| AIME 2024 | 13.1% ↗ | not reported | — | — |
| SWE-bench Verified (Epoch AI run) | 31% ↗epoch | not reported | — | — |
| Only Kimi K3 reports · 53 | ||||
| AIME 2026 | not reported | 96.7% ↗matharena ⚠ | — | — |
| Humanity's Last Exam (no tools) | not reported | 43.5% ↗ | — | — |
| AA-Briefcase | not reported | 1548 ↗ | — | — |
| AA-LCR | not reported | 74.7% ↗ | — | — |
| Agents' Last Exam | not reported | 28.3% ↗ | — | — |
| APEX-Agents | not reported | 41% ↗ | — | — |
| AutomationBench | not reported | 30.8% ↗ | — | — |
| BabyVision (with Python) | not reported | 85.7% ↗ | — | — |
| BrowseComp (1M context, no compaction) | not reported | 90.4% ↗ | — | — |
| BrowseComp (context compaction) | not reported | 91.2% ↗ | — | — |
| CharXiv Reasoning (no tools) | not reported | 84.8% ↗ | — | — |
| CharXiv Reasoning (with Python) | not reported | 91.3% ↗ | — | — |
| CorpFin v2 | not reported | 71.6% ↗ | — | — |
| CritPt | not reported | 23.4% ↗ | — | — |
| DeepSearchQA (F1) | not reported | 95% ↗ | — | — |
| DeepSWE v1.1 | not reported | 67.3% ↗ | — | — |
| DeepSWE v1.1 (Kimi Code) | not reported | 67.5% ↗ | — | — |
| Finance Agent v2 | not reported | 54.4% ↗ | — | — |
| FrontierMath Tier 4 v2 (Epoch AI run) | not reported | 39% ↗epoch | — | — |
| FrontierSWE | not reported | 81.2% ↗ | — | — |
| GDPval-AA v2 | not reported | 1686 ↗ | — | — |
| Harvey Lab-AA | not reported | 94.6% ↗ | — | — |
| Humanity's Last Exam (with tools) | not reported | 56% ↗ | — | — |
| JobBench | not reported | 54.3% ↗ | — | — |
| Kimi Code Bench 2.0 | not reported | 72.9% ↗ | — | — |
| Legal Research Bench | not reported | 44.2% ↗ | — | — |
| MathVision (no tools) | not reported | 94.3% ↗ | — | — |
| MathVision (with Python) | not reported | 97.8% ↗ | — | — |
| MCP-Atlas | not reported | 84.2% ↗ | — | — |
| MCPMark-Verified | not reported | 94.5% ↗ | — | — |
| MLS-Bench-Lite | not reported | 48.3% ↗ | — | — |
| MMMU-Pro (no tools) | not reported | 81.6% ↗ | — | — |
| MMMU-Pro (with Python) | not reported | 83.4% ↗ | — | — |
| MMVU | not reported | 82.1% ↗ | — | — |
| OfficeQA Pro | not reported | 63.3% ↗ | — | — |
| OmniDocBench | not reported | 91.1% ↗ | — | — |
| OSWorld 2.0 | not reported | 58.3% ↗ | — | — |
| OSWorld-Verified | not reported | 84.8% ↗ | — | — |
| PerceptionBench | not reported | 58.5% ↗ | — | — |
| PostTrainBench | not reported | 36.6% ↗ | — | — |
| ProgramBench | not reported | 77.8% ↗ | — | — |
| ResearchRubrics | not reported | 76.2% ↗ | — | — |
| SaaS-Bench | not reported | 60.1% ↗ | — | — |
| SciCode | not reported | 58.7% ↗ | — | — |
| SpreadsheetBench 2 | not reported | 34.8% ↗ | — | — |
| SWE-Marathon | not reported | 42% ↗ | — | — |
| tau3-Banking | not reported | 33.4% ↗ | — | — |
| Terminal-Bench 2.1 | not reported | 88.3% ↗ | — | — |
| Toolathlon-Verified | not reported | 76.5% ↗ | — | — |
| Video-MME (with subtitles) | not reported | 90% ↗ | — | — |
| WorldVQA ForceAnswer | not reported | 51% ↗ | — | — |
| ZeroBench pass@5 (no tools) | not reported | 23% ↗ | — | — |
| ZeroBench pass@5 (with Python) | not reported | 41% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is GPT-4o minus Kimi K3 in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | GPT-4o | Kimi K3 | Edge |
|---|---|---|---|
| Context window | 128K tokens | 1M tokens | Kimi K3 |
| Max output | 16K tokens | 1.0M tokens | Kimi K3 |
| Input price / 1M | $2.5 ↗ | $3 ↗ | GPT-4o |
| Output price / 1M | $10 ↗ | $15 ↗ | GPT-4o |
| Cached input / 1M | $1.25 ↗ | $0.3 ↗ | Kimi K3 |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $12.5 | $18 | GPT-4o |
| Modalities | text · vision | text · vision | Tie |
| Released | 2024-05-13 | 2026-07-16 | — |
| Cited benchmark scores | 9 | 59 | — |
Reliability
Provider status
Last 60 days · refreshed daily on this page · live on /statusMore matchups
GPT-4o vs …
Models sharing the most benchmarksMore matchups
Kimi K3 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run GPT-4o and Kimi K3 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.