deepseek-v3.2 vs kimi-k2-0905-preview
deepseek-v3.2 is 2.1× cheaper on input than kimi-k2-0905-preview.
| deepseek-v3.2 | kimi-k2-0905-preview | |
|---|---|---|
| Input / 1M | $0.28 | $0.60 |
| Cached input / 1M | — | $0.15 |
| Output / 1M | $0.40 | $2.50 |
| Context window | 164K | 262K |
| Max output | 164K | 262K |
| Provider | DeepSeek | Moonshot |
| Vision | No | No |
| Tool calling | Yes | Yes |
| Token count | estimated | estimated |
Capability for the money
Independent measurement · Artificial Analysis · read 2026-08-11deepseek-v3.2 is 3.5× cheaper than kimi-k2-0905-preview and scores higher on following instructions (IFBench) — there is no case for paying more here.
Following instructions
extraction, classification, sticking to a format
IFBenchdeepseek-v3.2 +7.3
Coding
writing and fixing code
LiveCodeBenchkimi-k2-0905-preview +1.7
Tool use
multi-step agent work against real tools
τ²-benchdeepseek-v3.2 +5.5
Hard reasoning
graduate-level science and multi-step analysis
GPQA Diamondkimi-k2-0905-preview +1.6
Averaged over the benchmarks both models sat, deepseek-v3.2 returns 211.5 points of pass rate per dollar of blended price against 58.8 — the ratio that decides whether the dearer model earns its rate.
Each figure is a pass rate: the share of that test set the model answered correctly, measured by a third party rather than self-reported by the lab. Rows are different tests of different difficulty, so compare a model against the other model, not one row against the next. And a benchmark is not your workload — it narrows the shortlist, it does not pick for you.
Which to pick
- Cheaper input: deepseek-v3.2
- Cheaper output: deepseek-v3.2
- Larger context: kimi-k2-0905-preview
- Better at following instructions: deepseek-v3.2 (49.0% vs 41.7% on IFBench)
- Better at coding: kimi-k2-0905-preview (59.3% vs 61.0% on LiveCodeBench)
- Better at tool use: deepseek-v3.2 (78.9% vs 73.4% on τ²-bench)
- Better at hard reasoning: kimi-k2-0905-preview (75.1% vs 76.7% on GPQA Diamond)
List price is not production cost. Output length, cache hit rate, retries and task quality move the real number more than the headline rate does — price your own prompt on the token ledger before deciding.
Related comparisons
deepseek-v3.2 detail · kimi-k2-0905-preview detail · All DeepSeek pricing · All comparisons