Compare / head-to-head
DeepSeek V3.2-Expvs
Kimi K2.5
Kimi K2.5 leads 8 of 8 shared benchmarks. Kimi K2.5 has the larger context window (256K tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
0 – 8
Kimi K2.5 leads
Cheaper per token
—
list price, input + output
Larger context
Kimi K2.5
256K tokens
Providers
2 providers
DeepSeek · Moonshot AI
DeepSeek V3.2-Exp
DeepSeek · released 2025-09-29
text
- Context
- 128K
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 17 · 7 core
Kimi K2.5
Moonshot AI · released 2026-01-27
textvisionvideo
- Context
- 256K
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 26 · 14 core
Quality
Benchmark matrix
8 shared · 9 only DeepSeek V3.2-Exp · 18 only Kimi K2.5| Benchmark | DeepSeek V3.2-Exp | Kimi K2.5 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 8 | ||||
| MMLU-Pro | 85% ↗ | 87.1% ↗ | -2.1 pt | Kimi K2.5 |
| AIME 2025 | 89.3% ↗ | 96.1% ↗ | -6.8 pt | Kimi K2.5 |
| LMArena Elo | 1424.9 ↗ | 1450.1 ↗ | -25.2 | Kimi K2.5 |
| BrowseComp | 40.1% ↗ | 60.6% ↗ | -20.5 pt | Kimi K2.5 |
| GPQA-Diamond | 79.9% ↗ | 87.6% ↗ | -7.7 pt | Kimi K2.5 |
| LMArena WebDev (LMArena) | 1271.8 ↗ | 1435.8 ↗ | -164 | Kimi K2.5 |
| Vending-Bench 2 (Andon Labs) | 1034 ↗ | 1198.46 ↗ | -164.5 | Kimi K2.5 |
| WeirdML (Håvard Tveit Ihle) | 39.5% ↗ | 45.6% ↗ | -6.1 pt | Kimi K2.5 |
| Only DeepSeek V3.2-Exp reports · 9 | ||||
| Humanity's Last Exam (no tools) | 19.8% ↗ | not reported | — | — |
| Aider-Polyglot | 74.5% ↗ | not reported | — | — |
| BrowseComp-zh | 47.9% ↗ | not reported | — | — |
| HMMT 2025 | 83.6% ↗ | not reported | — | — |
| LiveCodeBench | 74.1% ↗ | not reported | — | — |
| SimpleQA (agentic tool use) | 97.1% ↗ | not reported | — | — |
| SWE Verified | 67.8% ↗ | not reported | — | — |
| SWE-bench Multilingual | 57.9% ↗ | not reported | — | — |
| Terminal-bench | 37.7% ↗ | not reported | — | — |
| Only Kimi K2.5 reports · 18 | ||||
| SWE-bench Verified | not reported | 76.8% ↗ | — | — |
| HLE-Full | not reported | 30.1% ↗ | — | — |
| HLE-Full (w/ tools) | not reported | 50.2% ↗ | — | — |
| HMMT 2025 (Feb) | not reported | 95.4% ↗ | — | — |
| IMO-AnswerBench | not reported | 81.8% ↗ | — | — |
| LiveCodeBench (v6) | not reported | 85% ↗ | — | — |
| LMArena Vision (LMArena) | not reported | 1252.2 ↗ | — | — |
| MMMU-Pro | not reported | 78.5% ↗ | — | — |
| OSWorld-Verified (XLANG) | not reported | 63.3% ↗ | — | — |
| SAGE (Vals AI) | not reported | 49.9% ↗ | — | — |
| SimpleBench (SimpleBench) | not reported | 46.8% ↗ | — | — |
| SimpleQA Verified | not reported | 34.3% ↗epoch run | — | — |
| SWE-Bench Multilingual | not reported | 73% ↗ | — | — |
| SWE-Bench Pro | not reported | 50.7% ↗ | — | — |
| SWE-bench Verified (Epoch AI run) | not reported | 73.8% ↗epoch run | — | — |
| Terminal Bench 2.0 | not reported | 50.8% ↗ | — | — |
| Toolathlon-Verified (HKUST) | not reported | 33% ↗ | — | — |
| VideoMMMU | not reported | 86.6% ↗ | — | — |
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 DeepSeek V3.2-Exp minus Kimi K2.5 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 | DeepSeek V3.2-Exp | Kimi K2.5 | Edge |
|---|---|---|---|
| Context window | 128K tokens | 256K tokens | Kimi K2.5 |
| Max output | — | — | — |
| Input price / 1M | — | — | — |
| Output price / 1M | — | — | — |
| Cached input / 1M | — | — | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | — | — | — |
| Modalities | text | text · vision · video | Kimi K2.5 |
| Released | 2025-09-29 | 2026-01-27 | — |
| Cited benchmark scores | 17 | 26 | — |
Reliability
Provider status
Live from /statusMore matchups
DeepSeek V3.2-Exp vs …
Models sharing the most benchmarksMore matchups
Kimi K2.5 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run DeepSeek V3.2-Exp and Kimi K2.5 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.