Model comparison
DeepSeek V4 Flash vs Kimi K2.5
DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 48.1 on the Noometry Index.
Last verified . 34 shared benchmarks.
DeepSeekSummary
- They share 34 benchmarks with published results for both. DeepSeek V4 Flash scores higher in 3 categories and Kimi K2.5 in 5 categories; 5 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek V4 Flash leads 53.7 to 31.2.
- The biggest single-benchmark swing is ARC-AGI-2: 61.4% for DeepSeek V4 Flash and 11.8% for Kimi K2.5.
- DeepSeek V4 Flash is cheaper at $0.15 / $0.60 per million input/output tokens, against $0.45 / $2.25 for Kimi K2.5.
- DeepSeek V4 Flash accepts more context: 1M tokens versus 262K.
Side by side
| DeepSeek V4 Flash | Kimi K2.5 | |
|---|---|---|
| Provider | DeepSeek | Moonshot AI |
| Noometry Index | 53.6 | 48.1 |
| Released | 2026-04-24 | 2026-01-27 |
| Weights | Open | Open |
| Context window | 1M | 262K |
| Max output | 393K | 262K |
| Input $ / M tokens | $0.15 | $0.45 |
| Output $ / M tokens | $0.60 | $2.25 |
| Results tracked | 41 | 51 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Too close to call
DeepSeek V4 Flash: 47.9 (#59), Kimi K2.5: 48.8 (#53)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| LMArena WebDev | 1582 | 1437 |
| SciCode | 49.9% | 49% |
| WeirdML | 63% | 45.6% |
| LMArena Coding | 1457 | 1474 |
| ALE-Bench | 1,306 | 821.65 |
| SWE-bench Verified | — | 73.8% |
| FrontierCode | 18.8% | — |
| SWE-bench Verified (bash only) | — | 70.8% |
| SWE-bench Multilingual | — | 67.3% |
Agentic & Tool Use Not comparable
DeepSeek V4 Flash: —, Kimi K2.5: 34.2 (#48)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| Terminal-Bench | — | 43.2% |
| OSWorld | — | 63.3% |
| Vending-Bench 2 | — | 1,198 |
Reasoning DeepSeek V4 Flash leads
DeepSeek V4 Flash: 53.7 (#30), Kimi K2.5: 31.2 (#80)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| ARC-AGI-2 | 61.4% | 11.8% |
| SimpleBench | 61.1% | 46.8% |
| Kagi LLM Benchmark | 52.2% | 78.5% |
| NYT Connections (extended) | 89.6% | 69.9% |
| ARC-AGI-1 | 89% | 65.3% |
| CritPt | 16.6% | 3.1% |
| Chess Puzzles | 33% | 12% |
| LMArena Hard Prompts | 1444 | 1453 |
| Epoch Capabilities Index | 154.49 | 148.03 |
| EnigmaEval | — | 3.4% |
| Thematic Generalization | — | 69.4% |
| Mystery Game Puzzles | 34% | — |
| DTBench | 90.9% | — |
| LMCA | 41.7% | — |
Math DeepSeek V4 Flash leads
DeepSeek V4 Flash: 60.3 (#37), Kimi K2.5: 51.8 (#53)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| MathArena Final-Answer Competitions | 76.5% | 62.3% |
| OTIS Mock AIME 2024-2025 | 94.4% | 92.2% |
| LMArena Math | 1427 | 1470 |
| FrontierMath (Tiers 1-3) | 57.5% | — |
| FrontierMath Tier 4 | 24.4% | — |
| ProofBench | 56% | — |
| FrontierMath (Feb 2025 set) | — | 27.9% |
| FrontierMath Tier 4 (v1) | — | 4.2% |
Knowledge DeepSeek V4 Flash leads
DeepSeek V4 Flash: 55.4 (#48), Kimi K2.5: 53.6 (#56)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| GPQA Diamond | 91% | 87.6% |
| SimpleQA Verified | 33.6% | 34.3% |
| LMArena Expert | 1441 | 1466 |
| Humanity's Last Exam | — | 24.4% |
| Vectara Hallucination Rate | — | 14.2% |
Multimodal Not comparable
DeepSeek V4 Flash: —, Kimi K2.5: 41.1 (#39)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| LMArena Vision | — | 1269 |
| LMArena Document | — | 1430 |
Multilingual Too close to call
DeepSeek V4 Flash: 53.0 (#72), Kimi K2.5: 53.9 (#53)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| LMArena Non-English | 1420 | 1433 |
| LMArena Chinese | 1468 | 1495 |
| LMArena French | 1439 | 1454 |
| LMArena German | 1418 | 1441 |
| LMArena Japanese | 1406 | 1421 |
| LMArena Korean | 1384 | 1410 |
| LMArena Russian | 1428 | 1435 |
| LMArena Spanish | 1436 | 1450 |
Instruction Following Too close to call
DeepSeek V4 Flash: 74.9 (#81), Kimi K2.5: 75.3 (#64)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| LMArena Instruction Following | 1421 | 1431 |
Long Context Kimi K2.5 leads
DeepSeek V4 Flash: 43.8 (#85), Kimi K2.5: 52.1 (#7)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| LMArena Longer Query | 1434 | 1445 |
| Fiction.LiveBench | — | 86.1% |
| CL-bench | — | 19.3% |
| CL-bench Life | — | 13.2% |
Writing & Preference Kimi K2.5 leads
DeepSeek V4 Flash: 63.8 (#61), Kimi K2.5: 65.1 (#53)
| Benchmark | DeepSeek V4 Flash | Kimi K2.5 |
|---|---|---|
| LMArena Text | 1432 | 1445 |
| LMArena Creative Writing | 1403 | 1423 |
| EQ-Bench Creative Writing | 1559 | 1579 |
| LMArena Multi-Turn | 1449 | 1444 |
Frequently asked questions
Is DeepSeek V4 Flash better than Kimi K2.5?
DeepSeek V4 Flash is the stronger model overall, scoring 53.6 to 48.1 on the Noometry Index.
Which is cheaper, DeepSeek V4 Flash or Kimi K2.5?
DeepSeek V4 Flash is cheaper. It lists at $0.15 per million input tokens and $0.60 per million output tokens; Kimi K2.5 lists at $0.45 and $2.25.
Is DeepSeek V4 Flash or Kimi K2.5 better for coding?
They score almost the same on coding (47.9 vs 48.8); test both on your own repository before choosing.
Which has the bigger context window?
DeepSeek V4 Flash does, with 1M tokens against 262K.
How many benchmarks do DeepSeek V4 Flash and Kimi K2.5 share?
34 benchmarks have published results for both models. DeepSeek V4 Flash has 41 scored results on Noometry and Kimi K2.5 has 51.