Model comparison
DeepSeek-V3.1 vs Kimi K2.5
Kimi K2.5 is the stronger model overall, scoring 48.1 to 42.8 on the Noometry Index. DeepSeek-V3.1 costs 2.1× less per token, which makes it the better buy when Kimi K2.5's lead doesn't matter for your workload.
Last verified . 24 shared benchmarks.
DeepSeekSummary
- They share 24 benchmarks with published results for both. DeepSeek-V3.1 scores higher in 0 categories and Kimi K2.5 in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in long context, where Kimi K2.5 leads 52.1 to 36.3.
- The biggest single-benchmark swing is Fiction.LiveBench: 52.8% for DeepSeek-V3.1 and 86.1% for Kimi K2.5.
- DeepSeek-V3.1 is cheaper at $0.25 / $0.95 per million input/output tokens, against $0.45 / $2.25 for Kimi K2.5.
- Kimi K2.5 accepts more context: 262K tokens versus 164K.
Side by side
| DeepSeek-V3.1 | Kimi K2.5 | |
|---|---|---|
| Provider | DeepSeek | Moonshot AI |
| Noometry Index | 42.8 | 48.1 |
| Released | 2025-08-21 | 2026-01-27 |
| Weights | Open | Open |
| Context window | 164K | 262K |
| Max output | 8K | 262K |
| Input $ / M tokens | $0.25 | $0.45 |
| Output $ / M tokens | $0.95 | $2.25 |
| Results tracked | 27 | 51 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Kimi K2.5 leads
DeepSeek-V3.1: 40.3 (#144), Kimi K2.5: 48.8 (#53)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| WeirdML | 38.4% | 45.6% |
| LMArena Coding | 1417 | 1474 |
| SWE-bench Verified | — | 73.8% |
| SWE-bench Verified (bash only) | — | 70.8% |
| LMArena WebDev | — | 1437 |
| SWE-bench Multilingual | — | 67.3% |
| SciCode | — | 49% |
| ALE-Bench | — | 821.65 |
Agentic & Tool Use Not comparable
DeepSeek-V3.1: —, Kimi K2.5: 34.2 (#48)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| Terminal-Bench | — | 43.2% |
| OSWorld | — | 63.3% |
| Vending-Bench 2 | — | 1,198 |
Reasoning Kimi K2.5 leads
DeepSeek-V3.1: 27.9 (#110), Kimi K2.5: 31.2 (#80)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| SimpleBench | 40% | 46.8% |
| Kagi LLM Benchmark | 53.2% | 78.5% |
| LMArena Hard Prompts | 1417 | 1453 |
| Epoch Capabilities Index | 139.92 | 148.03 |
| ARC-AGI-2 | — | 11.8% |
| NYT Connections (extended) | — | 69.9% |
| ARC-AGI-1 | — | 65.3% |
| CritPt | — | 3.1% |
| Chess Puzzles | — | 12% |
| EnigmaEval | — | 3.4% |
| Thematic Generalization | — | 69.4% |
| DTBench | 82.7% | — |
| LMCA | 24.3% | — |
| ForecastBench | 58 | — |
Math Kimi K2.5 leads
DeepSeek-V3.1: 38.9 (#122), Kimi K2.5: 51.8 (#53)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| LMArena Math | 1420 | 1470 |
| MathArena Final-Answer Competitions | — | 62.3% |
| OTIS Mock AIME 2024-2025 | — | 92.2% |
| FrontierMath (Feb 2025 set) | — | 27.9% |
| FrontierMath Tier 4 (v1) | — | 4.2% |
Knowledge Kimi K2.5 leads
DeepSeek-V3.1: 43.7 (#90), Kimi K2.5: 53.6 (#56)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| Vectara Hallucination Rate | 5.5% | 14.2% |
| LMArena Expert | 1405 | 1466 |
| GPQA Diamond | — | 87.6% |
| Humanity's Last Exam | — | 24.4% |
| SimpleQA Verified | — | 34.3% |
Multimodal Not comparable
DeepSeek-V3.1: —, Kimi K2.5: 41.1 (#39)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| LMArena Vision | — | 1269 |
| LMArena Document | — | 1430 |
Multilingual Kimi K2.5 leads
DeepSeek-V3.1: 51.6 (#106), Kimi K2.5: 53.9 (#53)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| LMArena Non-English | 1400 | 1433 |
| LMArena Chinese | 1469 | 1495 |
| LMArena French | 1447 | 1454 |
| LMArena German | 1411 | 1441 |
| LMArena Japanese | 1378 | 1421 |
| LMArena Korean | 1337 | 1410 |
| LMArena Russian | 1405 | 1435 |
| LMArena Spanish | 1431 | 1450 |
Instruction Following Kimi K2.5 leads
DeepSeek-V3.1: 73.9 (#110), Kimi K2.5: 75.3 (#64)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| LMArena Instruction Following | 1400 | 1431 |
Long Context Kimi K2.5 leads
DeepSeek-V3.1: 36.3 (#232), Kimi K2.5: 52.1 (#7)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| Fiction.LiveBench | 52.8% | 86.1% |
| LMArena Longer Query | 1422 | 1445 |
| CL-bench | — | 19.3% |
| CL-bench Life | — | 13.2% |
Writing & Preference Kimi K2.5 leads
DeepSeek-V3.1: 60.3 (#98), Kimi K2.5: 65.1 (#53)
| Benchmark | DeepSeek-V3.1 | Kimi K2.5 |
|---|---|---|
| LMArena Text | 1420 | 1445 |
| LMArena Creative Writing | 1401 | 1423 |
| EQ-Bench Creative Writing | 1436 | 1579 |
| LMArena Multi-Turn | 1408 | 1444 |
Frequently asked questions
Is DeepSeek-V3.1 better than Kimi K2.5?
Kimi K2.5 is the stronger model overall, scoring 48.1 to 42.8 on the Noometry Index. DeepSeek-V3.1 costs 2.1× less per token, which makes it the better buy when Kimi K2.5's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3.1 or Kimi K2.5?
DeepSeek-V3.1 is cheaper. It lists at $0.25 per million input tokens and $0.95 per million output tokens; Kimi K2.5 lists at $0.45 and $2.25.
Is DeepSeek-V3.1 or Kimi K2.5 better for coding?
Kimi K2.5 scores higher on coding benchmarks: 48.8 versus 40.3 in the Noometry coding category.
Which has the bigger context window?
Kimi K2.5 does, with 262K tokens against 164K.
How many benchmarks do DeepSeek-V3.1 and Kimi K2.5 share?
24 benchmarks have published results for both models. DeepSeek-V3.1 has 27 scored results on Noometry and Kimi K2.5 has 51.