Model comparison
Kimi K2 Thinking vs DeepSeek-V3.1
- Input
- $0.60 / 1M tokens
- Output
- $2.50 / 1M tokens
- Context
- 256K
- Released
- 2025-11-06
- Input
- $0.21 / 1M tokens
- Output
- $0.79 / 1M tokens
- Context
- 128K
- Released
- 2025-08-21
Kimi K2 Thinking scores higher on 5 of the 6 benchmarks both models report (the other leads on 1); DeepSeek-V3.1 is cheaper on input tokens ($0.2 vs $0.6 per 1M, 2.9×); Kimi K2 Thinking has the larger context window (256K vs 128K).
Specs and pricing side by side
| Spec | Kimi K2 Thinking | DeepSeek-V3.1 |
|---|---|---|
| Maker | Moonshot AI (Kimi) | DeepSeek |
| Released | 2025-11-06 | 2025-08-21 |
| Context window | 256K | 128K |
| Max output | 256K | not published |
| Input price | $0.60 / 1M tokens | $0.21 / 1M tokens |
| Output price | $2.50 / 1M tokens | $0.79 / 1M tokens |
| Parameters | 1T total / 32B active (MoE) | 671B total / 37B active (MoE) |
| License | Modified MIT License | MIT |
| Weights | Open weights | Open weightssame |
| Modalities | Text | Textsame |
Not published by either maker: knowledge cutoff.
Benchmarks both models report
| Benchmark | Kimi K2 Thinking | DeepSeek-V3.1 |
|---|---|---|
| Humanity's Last Exam | 44.9% | 15.9% |
| BrowseComp | 60.2% | 30% |
| SWE-bench Verified | 71.3% | 66% |
| AIME 2025 | 99.1% | 88.4% |
| HMMT 2025 | 95.1% | 84.2% |
| MMLU-Pro | 84.6% | 84.8% |
Scores are as published by each maker or its cited source; harness, reasoning effort and date can differ between rows, so treat small gaps as ties. Sources are listed on each model page.
Which one should you pick?
Pick Kimi K2 Thinking if…
- You care about Humanity's Last Exam, BrowseComp, SWE-bench Verified (it scores higher on 5 shared benchmarks).
- You need the longer context (256K vs 128K).
- Your use case is: autonomous research agents that browse, gather, and synthesize across many steps.
Pick DeepSeek-V3.1 if…
- You care about MMLU-Pro (it scores higher on 1 shared benchmark).
- Cost matters: input tokens cost 2.9× less ($0.2 vs $0.6 per 1M).
- Your use case is: coding agents and software-engineering tasks (SWE-bench-style fix/patch workflows).
FAQ
Which is better, Kimi K2 Thinking or DeepSeek-V3.1?
Kimi K2 Thinking scores higher on 5 of the 6 benchmarks both models report (the other leads on 1); DeepSeek-V3.1 is cheaper on input tokens ($0.2 vs $0.6 per 1M, 2.9×); Kimi K2 Thinking has the larger context window (256K vs 128K). The better pick depends on whether you value benchmark scores, price per token, context length or openness for your workload.
How much cheaper is DeepSeek-V3.1?
Kimi K2 Thinking: $0.60 / 1M tokens input, $2.50 / 1M tokens output. DeepSeek-V3.1: $0.21 / 1M tokens input, $0.79 / 1M tokens output (list API prices as published by each maker).
What are the context windows of Kimi K2 Thinking and DeepSeek-V3.1?
Kimi K2 Thinking offers 256K with up to 256K output; DeepSeek-V3.1 offers 128K.
Is Kimi K2 Thinking newer than DeepSeek-V3.1?
Kimi K2 Thinking was released 2025-11-06; DeepSeek-V3.1 was released 2025-08-21.
Go deeper
- Kimi K2 Thinking: full specs, pricing and benchmarks
- DeepSeek-V3.1: full specs, pricing and benchmarks
- deepseek v3 1 terminus vs deepseek v3 1
- glm 4 7 flash vs kimi k2 thinking
- deepseek v3 1 vs minimax m1
- kimi k2 thinking vs deepseek v3 2 exp
- minimax m2 1 vs deepseek v3 1
- kimi k2 thinking vs deepseek v3 1 terminus
- All model comparisons