Model comparison
DeepSeek V4 Pro vs Kimi K3
Kimi K3 is the stronger model overall, scoring 59.5 to 54.3 on the Noometry Index. DeepSeek V4 Pro costs 6.1× less per token, which makes it the better buy when Kimi K3's lead doesn't matter for your workload.
Last verified . 44 shared benchmarks.
DeepSeekSummary
- They share 44 benchmarks with published results for both. DeepSeek V4 Pro scores higher in 0 categories and Kimi K3 in 9 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where Kimi K3 leads 76.6 to 65.5.
- The biggest single-benchmark swing is Surface Evolver Bench: 40% for DeepSeek V4 Pro and 95% for Kimi K3.
- DeepSeek V4 Pro is cheaper at $0.66 / $1.98 per million input/output tokens, against $3 / $15 for Kimi K3.
- Kimi K3 accepts more context: 1.05M tokens versus 1M.
Side by side
| DeepSeek V4 Pro | Kimi K3 | |
|---|---|---|
| Provider | DeepSeek | Moonshot AI |
| Noometry Index | 54.3 | 59.5 |
| Released | 2026-04-24 | 2026-07-16 |
| Weights | Open | Open |
| Context window | 1M | 1.05M |
| Max output | 393K | 1.05M |
| Input $ / M tokens | $0.66 | $3 |
| Output $ / M tokens | $1.98 | $15 |
| Results tracked | 48 | 53 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Kimi K3 leads
DeepSeek V4 Pro: 52.4 (#34), Kimi K3: 61.0 (#10)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| FrontierCode | 28.6% | 44.2% |
| LMArena WebDev | 1582 | 1654 |
| SciCode | 51% | 59.5% |
| WeirdML | 66.2% | 82.6% |
| LMArena Coding | 1470 | 1508 |
| ALE-Bench | 1,403 | 1,524 |
| SWE-bench Verified | 77.6% | — |
| DeepSWE | — | 68.5% |
| FrontierSWE | — | 25.9% |
Agentic & Tool Use Kimi K3 leads
DeepSeek V4 Pro: 32.8 (#58), Kimi K3: 41.8 (#20)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| APEX-Agents | 47.3% | 50.6% |
| Vending-Bench 2 | 3,285 | 5,165 |
| τ²-bench Banking | — | 37.1% |
| PostTrainBench | — | 32% |
| GBAEval | — | 48.3% |
| GDP.pdf | — | 19% |
Reasoning Kimi K3 leads
DeepSeek V4 Pro: 56.5 (#24), Kimi K3: 63.0 (#17)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| ARC-AGI-2 | 61.3% | 60.4% |
| NYT Connections (extended) | 91.3% | 93.6% |
| ARC-AGI-1 | 90.5% | 94.5% |
| CritPt | 18% | 23.4% |
| Chess Puzzles | 47% | 39% |
| LMArena Hard Prompts | 1461 | 1496 |
| Mystery Game Puzzles | 43% | 26% |
| DTBench | 93.9% | 91.2% |
| LMCA | 45.5% | 52.7% |
| Surface Evolver Bench | 40% | 95% |
| Epoch Capabilities Index | 155.31 | 157.45 |
| ForecastBench | 56.1 | 61.1 |
| SimpleBench | — | 60.7% |
| Kagi LLM Benchmark | 53.5% | — |
Math Kimi K3 leads
DeepSeek V4 Pro: 64.8 (#30), Kimi K3: 74.2 (#16)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| FrontierMath (Tiers 1-3) | 64.6% | 72.2% |
| FrontierMath Tier 4 | 26.8% | 39% |
| MathArena Final-Answer Competitions | 76.6% | 87.8% |
| OTIS Mock AIME 2024-2025 | 98.6% | 97.2% |
| ProofBench | 50% | 87% |
| LMArena Math | 1455 | 1491 |
Knowledge Kimi K3 leads
DeepSeek V4 Pro: 59.5 (#31), Kimi K3: 63.2 (#21)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| GPQA Diamond | 91.7% | 93.1% |
| SimpleQA Verified | 52.9% | 50.6% |
| LMArena Expert | 1464 | 1521 |
| Vectara Hallucination Rate | 8.6% | — |
Multimodal Not comparable
DeepSeek V4 Pro: —, Kimi K3: 37.8 (#70)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| Blueprint-Bench 2 | — | 29.5% |
| Furniture Assembly | — | 34.2% |
Multilingual Kimi K3 leads
DeepSeek V4 Pro: 54.4 (#45), Kimi K3: 56.3 (#21)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| LMArena Non-English | 1439 | 1466 |
| LMArena Chinese | 1486 | 1529 |
| LMArena French | 1472 | 1491 |
| LMArena German | 1458 | 1488 |
| LMArena Japanese | 1445 | 1487 |
| LMArena Korean | 1447 | 1458 |
| LMArena Russian | 1453 | 1482 |
| LMArena Spanish | 1458 | 1472 |
Instruction Following Kimi K3 leads
DeepSeek V4 Pro: 76.1 (#47), Kimi K3: 77.7 (#14)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| LMArena Instruction Following | 1448 | 1483 |
Long Context Too close to call
DeepSeek V4 Pro: 45.0 (#51), Kimi K3: 45.8 (#29)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| LMArena Longer Query | 1458 | 1494 |
| CL-bench Life | 13.5% | — |
Writing & Preference Kimi K3 leads
DeepSeek V4 Pro: 65.5 (#46), Kimi K3: 76.6 (#4)
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| LMArena Text | 1451 | 1476 |
| LMArena Creative Writing | 1446 | 1454 |
| EQ-Bench Creative Writing | 1553 | 2082 |
| EQ-Bench 4 | 1166 | 1339 |
| LMArena Multi-Turn | 1467 | 1488 |
Frequently asked questions
Is DeepSeek V4 Pro better than Kimi K3?
Kimi K3 is the stronger model overall, scoring 59.5 to 54.3 on the Noometry Index. DeepSeek V4 Pro costs 6.1× less per token, which makes it the better buy when Kimi K3's lead doesn't matter for your workload.
Which is cheaper, DeepSeek V4 Pro or Kimi K3?
DeepSeek V4 Pro is cheaper. It lists at $0.66 per million input tokens and $1.98 per million output tokens; Kimi K3 lists at $3 and $15.
Is DeepSeek V4 Pro or Kimi K3 better for coding?
Kimi K3 scores higher on coding benchmarks: 61.0 versus 52.4 in the Noometry coding category.
Which has the bigger context window?
Kimi K3 does, with 1.05M tokens against 1M.
How many benchmarks do DeepSeek V4 Pro and Kimi K3 share?
44 benchmarks have published results for both models. DeepSeek V4 Pro has 48 scored results on Noometry and Kimi K3 has 53.