Model comparison
DeepSeek-V3 vs Kimi K2 (Jul 2025)
Kimi K2 (Jul 2025) is the stronger model overall, scoring 41.2 to 39.5 on the Noometry Index. DeepSeek-V3 costs 2.5× less per token, which makes it the better buy when Kimi K2 (Jul 2025)'s lead doesn't matter for your workload.
Last verified . 35 shared benchmarks.
DeepSeekSummary
- They share 35 benchmarks with published results for both. DeepSeek-V3 scores higher in 2 categories and Kimi K2 (Jul 2025) in 6 categories; 6 gaps are clear of the uncertainty.
- The widest gap is in math, where Kimi K2 (Jul 2025) leads 42.7 to 32.1.
- The biggest single-benchmark swing is Omni-MATH: 40.3% for DeepSeek-V3 and 65.4% for Kimi K2 (Jul 2025).
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $0.57 / $2.30 for Kimi K2 (Jul 2025).
- Kimi K2 (Jul 2025) accepts more context: 262K tokens versus 164K.
Side by side
| DeepSeek-V3 | Kimi K2 (Jul 2025) | |
|---|---|---|
| Provider | DeepSeek | Moonshot AI |
| Noometry Index | 39.5 | 41.2 |
| Released | 2024-12-26 | 2025-07-12 |
| Weights | Open | Open |
| Context window | 164K | 262K |
| Max output | 164K | 262K |
| Input $ / M tokens | $0.24 | $0.57 |
| Output $ / M tokens | $0.90 | $2.30 |
| Results tracked | 60 | 42 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
DeepSeek-V3: 42.3 (#106), Kimi K2 (Jul 2025): 42.4 (#102)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| Aider Polyglot | 55.1% | 59.1% |
| WeirdML | 36.1% | 42.8% |
| LMArena Coding | 1368 | 1399 |
| SWE-bench Verified (bash only) | — | 63.4% |
| SciCode | 35.8% | — |
| GSO | — | 4.9% |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| ALE-Bench | — | 597.5 |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |
Agentic & Tool Use Not comparable
DeepSeek-V3: —, Kimi K2 (Jul 2025): 32.4 (#64)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| METR Time Horizons | 49.6% | 59.2% |
| Terminal-Bench | — | 35.7% |
| Berkeley Function Calling Leaderboard | — | 59.1% |
Reasoning Kimi K2 (Jul 2025) leads
DeepSeek-V3: 20.5 (#236), Kimi K2 (Jul 2025): 23.3 (#179)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| SimpleBench | 27.2% | 26.3% |
| Kagi LLM Benchmark | 52.3% | 64.4% |
| LMArena Hard Prompts | 1365 | 1384 |
| Epoch Capabilities Index | 135.94 | 146.01 |
| ForecastBench | 59.1 | 60.2 |
| CritPt | 0% | — |
| LiveBench Reasoning | 65.8% | — |
| DTBench | 64.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| LMCA | 15.5% | — |
| BIG-Bench Hard | 87.5% | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |
Math Kimi K2 (Jul 2025) leads
DeepSeek-V3: 32.1 (#219), Kimi K2 (Jul 2025): 42.7 (#83)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| Omni-MATH | 40.3% | 65.4% |
| LMArena Math | 1373 | 1397 |
| FrontierMath (Feb 2025 set) | 1.7% | 21.4% |
| OTIS Mock AIME 2024-2025 | 37.8% | — |
| LiveBench Math | 73.5% | — |
| MATH Level 5 | 75.5% | — |
| FrontierMath Tier 4 (v1) | — | 0% |
Knowledge Too close to call
DeepSeek-V3: 37.5 (#155), Kimi K2 (Jul 2025): 37.3 (#157)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| MMLU-Pro | 72.3% | 81.9% |
| Confabulations | 26.1% | 20.4% |
| Vectara Hallucination Rate | 6.1% | 17.9% |
| GPQA (HELM) | 53.8% | 65.3% |
| LMArena Expert | 1351 | 1365 |
| GPQA Diamond | 67.6% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |
Multilingual Kimi K2 (Jul 2025) leads
DeepSeek-V3: 48.5 (#143), Kimi K2 (Jul 2025): 49.6 (#130)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| LMArena Non-English | 1358 | 1372 |
| LMArena Chinese | 1391 | 1415 |
| LMArena French | 1385 | 1379 |
| LMArena German | 1374 | 1387 |
| LMArena Japanese | 1333 | 1349 |
| LMArena Korean | 1319 | 1325 |
| LMArena Russian | 1373 | 1385 |
| LMArena Spanish | 1358 | 1386 |
Instruction Following DeepSeek-V3 leads
DeepSeek-V3: 72.8 (#130), Kimi K2 (Jul 2025): 71.1 (#156)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| IFEval | 83.2% | 85% |
| LMArena Instruction Following | 1345 | 1348 |
| LiveBench Instruction Following | 81.5% | — |
Long Context Kimi K2 (Jul 2025) leads
DeepSeek-V3: 34.0 (#253), Kimi K2 (Jul 2025): 41.2 (#145)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| Fiction.LiveBench | 50% | 66.7% |
| LMArena Longer Query | 1352 | 1353 |
| CL-bench | — | 17.6% |
Writing & Preference Kimi K2 (Jul 2025) leads
DeepSeek-V3: 57.4 (#130), Kimi K2 (Jul 2025): 62.3 (#78)
| Benchmark | DeepSeek-V3 | Kimi K2 (Jul 2025) |
|---|---|---|
| LMArena Text | 1375 | 1380 |
| LMArena Creative Writing | 1364 | 1350 |
| Short-Story Creative Writing | 77% | 85.6% |
| EQ-Bench Creative Writing | 1472 | 1666 |
| WildBench | 83% | 86.2% |
| LMArena Multi-Turn | 1389 | 1371 |
| LiveBench Language | 49.1% | — |
Frequently asked questions
Is DeepSeek-V3 better than Kimi K2 (Jul 2025)?
Kimi K2 (Jul 2025) is the stronger model overall, scoring 41.2 to 39.5 on the Noometry Index. DeepSeek-V3 costs 2.5× less per token, which makes it the better buy when Kimi K2 (Jul 2025)'s lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3 or Kimi K2 (Jul 2025)?
DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; Kimi K2 (Jul 2025) lists at $0.57 and $2.30.
Is DeepSeek-V3 or Kimi K2 (Jul 2025) better for coding?
They score almost the same on coding (42.3 vs 42.4); test both on your own repository before choosing.
Which has the bigger context window?
Kimi K2 (Jul 2025) does, with 262K tokens against 164K.
How many benchmarks do DeepSeek-V3 and Kimi K2 (Jul 2025) share?
35 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and Kimi K2 (Jul 2025) has 42.