Model comparison
DeepSeek V4 Pro vs Kimi K3
Head-to-head evidence from 6 shared benchmark results across 3 categories. Overall scores shown here use the public BenchAlign v5 ranking lane.
Public leaderboard positions: DeepSeek V4 Pro #46 (Supported); Kimi K3 #4 (Supported). Intervals and evidence labels describe ranking uncertainty, not a guarantee for a specific workload.
Evidence parity. DeepSeek V4 Pro and Kimi K3 share 6 comparable benchmark results. 2 of 8 categories are comparable. 17 results are unique to DeepSeek V4 Pro; 51 to Kimi K3.
Updated July 23, 2026- Shared results
- 6
- DeepSeek V4 Pro only
- 17
- Kimi K3 only
- 51
- Comparable categories
- 2 / 8
Pick Kimi K3 if you want the stronger benchmark profile. DeepSeek V4 Pro only becomes the better choice if you want the cheaper token bill or you would rather avoid the extra latency and token burn of a reasoning model.
Confidence note. This is a partial-evidence comparison with 6 shared benchmark results across 3 evidence categories; 2 of 8 categories currently have scoreable aggregates for both models. Treat the verdict as directional until coverage is more balanced.
Why this result
Kimi K3 is clearly ahead on the BenchAlign aggregate, 80.96 to 60.66. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Kimi K3's sharpest advantage is in agentic, where it averages 89.5 against 59.1. The single biggest benchmark swing on the page is HLE, 7.7% to 56%.
Kimi K3 is also the more expensive model on tokens at $3.00 input / $15.00 output per 1M tokens, versus $0.43 input / $0.87 output per 1M tokens for DeepSeek V4 Pro. That is roughly 17.2x on output cost alone. Kimi K3 is the reasoning model in the pair, while DeepSeek V4 Pro is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. Kimi K3 gives you the larger context window at 1.05M, compared with 1M for DeepSeek V4 Pro.
Category breakdown
Exact category averages are shown below. Not measured means BenchLM does not have enough sourced public coverage for that model and category.
| Category | DeepSeek V4 Pro | Δ | Kimi K3 |
|---|---|---|---|
| Agentic | DeepSeek V4 Pro59.1 | Margin→ 30.4 | Kimi K389.5 |
| Knowledge | DeepSeek V4 Pro41.3 | Margin→ 19.7 | Kimi K361.0 |
| Coding | DeepSeek V4 Pro65.3 | MarginNo overlap | Kimi K3Not measured |
| Math | DeepSeek V4 Pro31.7 | MarginNo overlap | Kimi K3Not measured |
| Multimodal | DeepSeek V4 ProNot measured | MarginNo overlap | Kimi K378.5 |
Decisive benchmark drivers
The largest measured benchmark gaps in this matchup, with exact reported values.
More
- Source ↗
HLE
KnowledgeA 7.7%B 56%Winner: Kimi K3Δ 48.3HLE: DeepSeek V4 Pro scored 7.7%; Kimi K3 scored 56%. Kimi K3 wins this benchmark. - Source ↗
Terminal-Bench 2.0
AgenticA 59.1%B 88.3%Winner: Kimi K3Δ 29.2Terminal-Bench 2.0: DeepSeek V4 Pro scored 59.1%; Kimi K3 scored 88.3%. Kimi K3 wins this benchmark. - Source ↗
GPQA
KnowledgeA 72.9%B 93.5%Winner: Kimi K3Δ 20.6GPQA: DeepSeek V4 Pro scored 72.9%; Kimi K3 scored 93.5%. Kimi K3 wins this benchmark.
Operational comparison
Runtime and commercial metrics are compared only when both models have a complete sourced value.
| Metric | DeepSeek V4 Pro | Kimi K3 | Comparison |
|---|---|---|---|
| Input / output priceUSD per 1M tokens | DeepSeek V4 Pro$0.435 input / $0.87 output | Kimi K3$3 input / $15 output | DeepSeek V4 Pro has the lower combined listed price. |
| Generation speedtokens per second | DeepSeek V4 ProNot available | Kimi K3Not available | A complete speed comparison is not available. |
| First-answer latencyseconds to first token | DeepSeek V4 ProNot available | Kimi K3Not available | A complete latency comparison is not available. |
| Context windowmaximum listed tokens | DeepSeek V4 Pro1M | Kimi K31.05M | Kimi K3 lists the larger context window. |
Benchmark Deep Dive
AgenticKimi K3 wins25 benchmarks
| Benchmark | DeepSeek V4 Pro | Kimi K3 | Result |
|---|---|---|---|
| Terminal-Bench 2.0Source | 59.1% | 88.3% | Kimi K3 leads |
| MCP AtlasSource | 69.4% | 84.2% | Kimi K3 leads |
| ToolathlonSource | 46.3% | — | Not comparable |
| Claw-EvalSource | 59.8% | — | Not comparable |
| Gert LabsSource | 50.28% | — | Not comparable |
| ResearchClawBenchSource | 17.1% | — | Not comparable |
| BrowseCompSource | — | 91.2% | Not comparable |
| DeepSearchQASource | — | 95.0% | Not comparable |
| Toolathlon-VerifiedSource | — | 73.2% | Not comparable |
| AutomationBenchSource | — | 30.8% | Not comparable |
| JobBenchSource | — | 52.9% | Not comparable |
| APEX-AgentsSource | — | 37.6% | Not comparable |
| SpreadsheetBench 2Source | — | 34.8% | Not comparable |
| DECK-BenchSource | — | 73.5% | Not comparable |
| AA Agentic IndexSource | — | 50.1% | Not comparable |
| GDPval-AASource | — | 59.0% | Not comparable |
| GDPval-AASource | — | 1679 | Not comparable |
| AA BriefcaseSource | — | 1543 | Not comparable |
| AA AutomationBenchSource | — | 52.7% | Not comparable |
| AA EnterpriseOps-GymSource | — | 45.3% | Not comparable |
| AA Harvey LABSource | — | 94.6% | Not comparable |
| AA Tau3 BankingSource | — | 33.4% | Not comparable |
| aaTerminalBench21Source | — | 85% | Not comparable |
| APEX-Agents-AASource | — | 41.3% | Not comparable |
| AA ITBenchSource | — | 47.7% | Not comparable |
Coding13 benchmarks
| Benchmark | DeepSeek V4 Pro | Kimi K3 | Result |
|---|---|---|---|
| SWE-bench VerifiedSource | 73.6% | — | Not comparable |
| SWE-bench ProSource | 52.1% | — | Not comparable |
| SWE MultilingualSource | 69.8% | — | Not comparable |
| Terminal-Bench 2.0Source | 59.1% | — | Not comparable |
| deepSweSource | — | 67.5% | Not comparable |
| FrontierSWESource | — | 81.2% | Not comparable |
| ProgramBenchSource | — | 77.8% | Not comparable |
| Kimi Code Bench v2Source | — | 72.9% | Not comparable |
| sweMarathonSource | — | 42% | Not comparable |
| PostTrain BenchSource | — | 36.6% | Not comparable |
| MLS-Bench LiteSource | — | 48.3% | Not comparable |
| AA Coding IndexSource | — | 76.2% | Not comparable |
| AA-SciCodeSource | — | 58.7% | Not comparable |
Reasoning4 benchmarks
KnowledgeKimi K3 wins13 benchmarks
| Benchmark | DeepSeek V4 Pro | Kimi K3 | Result |
|---|---|---|---|
| MMLU-ProSource | 82.9% | — | Not comparable |
| SimpleQASource | 45% | — | Not comparable |
| Chinese-SimpleQASource | 75.8% | — | Not comparable |
| GPQASource | 72.9% | 93.5% | Kimi K3 leads |
| GPQA-DSource | 72.9% | 93.5% | Kimi K3 leads |
| HLESource | 7.7% | 56% | Kimi K3 leads |
| HLE w/o toolsSource | — | 43.5% | Not comparable |
| Artificial Analysis Intelligence IndexSource | — | 57.1% | Not comparable |
| AA-GPQA DiamondSource | — | 93.5% | Not comparable |
| AA-HLESource | — | 44.3% | Not comparable |
| AA-Omniscience IndexSource | — | 18.4% | Not comparable |
| AA-Omniscience AccuracySource | — | 46.0% | Not comparable |
| AA-Omniscience Hallucination RateSource | — | 50.9% | Not comparable |
Math4 benchmarks
Multimodal15 benchmarks
| Benchmark | DeepSeek V4 Pro | Kimi K3 | Result |
|---|---|---|---|
| Design Arena WebsiteSource | 1264 | 1386 | Kimi K3 leads |
| OfficeQA ProSource | — | 63.3% | Not comparable |
| MMMU-ProSource | — | 81.6% | Not comparable |
| MMMU-Pro w/ PythonSource | — | 83.4% | Not comparable |
| CharXiv w/o toolsSource | — | 84.8% | Not comparable |
| CharXivSource | — | 91.3% | Not comparable |
| MathVisionSource | — | 94.3% | Not comparable |
| MathVision w/ PythonSource | — | 97.8% | Not comparable |
| BabyVision w/ PythonSource | — | 85.7% | Not comparable |
| ZeroBenchSource | — | 23.0% | Not comparable |
| ZeroBench w/ PythonSource | — | 41.0% | Not comparable |
| WorldVQA ForceAnswerSource | — | 51.0% | Not comparable |
| OmniDocBenchSource | — | 91.1% | Not comparable |
| PerceptionBenchSource | — | 58.5% | Not comparable |
| AA-MMMU-ProSource | — | 80.5% | Not comparable |
Frequently Asked Questions (3)
Which is better, DeepSeek V4 Pro or Kimi K3?
Kimi K3 is ahead on BenchLM's BenchAlign leaderboard, 80.96 to 60.66. The biggest single separator in this matchup is HLE, where the scores are 7.7% and 56%.
Which is better for knowledge tasks, DeepSeek V4 Pro or Kimi K3?
Kimi K3 has the edge for knowledge tasks in this comparison, averaging 61 versus 41.3. Inside this category, HLE is the benchmark that creates the most daylight between them.
Which is better for agentic tasks, DeepSeek V4 Pro or Kimi K3?
Kimi K3 has the edge for agentic tasks in this comparison, averaging 89.5 versus 59.1. Inside this category, Terminal-Bench 2.0 is the benchmark that creates the most daylight between them.
Related Comparisons
Explore More
Choose a model with this week’s evidence
Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.
One email each week. Unsubscribe anytime.