DeepSeek V4 Pro vs Kimi K3
| Spec | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| Respan Index | 49.2 (#45) | 52.8 (#35) |
| Blended price per 1M | $0.585 | $2.52 |
| Context window | 1M | 1.0M |
| Input / output per 1M | $0.28 / $1.5 | $0.192 / $9.5 |
| Max output | 384K | 1.0M |
| Input | text | text, vision |
| Released | 2026-08-13 | 2026-07-16 |
Respan Index by category
Each category is scored 0 to 100, and how far ahead the winner is. Leads under a point are close to even.
| Category | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| Coding | 29.8 DeepSeek V4 Pro | 38.8 Kimi K3 |
| Agents | 59.7 DeepSeek V4 Pro | 61.8 Kimi K3 |
| Reasoning | 52.8 DeepSeek V4 Pro | 54.8 Kimi K3 |
| Knowledge & language | 52.6 DeepSeek V4 Pro | 52.9 Kimi K3 |
| Human preference | 55.6 DeepSeek V4 Pro | 57.7 Kimi K3 |
Benchmarks both report
Kimi K3 leads on 20 of the 28 benchmarks both report. Higher is better. Each score links to where it was published.
| Benchmark | DeepSeek V4 Pro | Kimi K3 |
|---|---|---|
| LMArena WebDev (LMArena) | 1581 | 1659 |
| LMArena Elo | 1464 | 1488 |
| LMArena Agent (LMArena) | -0.003 | -0.0015 |
| MirrorCode (Epoch AI) | 33.3% | 54.4% |
| Chess Puzzles (Epoch AI) | 46% | 37% |
| APEX-Agents (Mercor) | 47.3% | 47.8% |
| DeepSWE v1.1 (Datacurve) | 62.8% | 68.5% |
| FrontierMath Tier 4 v2 (Epoch AI) | 26.8% | 39% |
| FrontierMath Tiers 1-3 v2 (Epoch AI) | 64.6% | 72.2% |
| GPQA Diamond | 91.7% | 93.5% |
| LiveBench Agentic Coding (LiveBench) | 54.9% | 62.2% |
| LiveBench Coding (LiveBench) | 77.2% | 81.4% |
| LiveBench Data Analysis (LiveBench) | 79.2% | 78.7% |
| LiveBench Instruction Following (LiveBench) | 67.7% | 71.4% |
| LiveBench Language (LiveBench) | 82.1% | 85.5% |
| LiveBench Mathematics (LiveBench) | 95.1% | 84.4% |
| LiveBench Reasoning (LiveBench) | 85.8% | 90.7% |
| Mystery Game Puzzles (Epoch AI) | 43% | 26% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 98.6% | 97.2% |
| SimpleQA Verified | 52.9% | 50.6% |
| Terminal-Bench 4.0 (Vals AI) | 14.1% | 17.2% |
| Vending-Bench 2 (Andon Labs) | 3284.52 | 5165.04 |
| WeirdML (Håvard Tveit Ihle) | 66.2% | 82.6% |
| AIME 2026 | 96.7% | 96.7% |
| Humanity's Last Exam (with tools) | 60% | 56% |
| Terminal-Bench 2.1 | 87.9% | 88.3% |
| Toolathlon-Verified (HKUST) | 74.4% | 76.5% |
| SimpleBench (SimpleBench) | 50.9% | 60.7% |
FAQ
Is DeepSeek V4 Pro better than Kimi K3?
Kimi K3 scores 52.8 on the Respan Index (#35), ahead of DeepSeek V4 Pro at 49.2 (#45). Kimi K3 leads on 20 of the 28 benchmarks both report.
Which is cheaper, DeepSeek V4 Pro or Kimi K3?
DeepSeek V4 Pro lists at $0.28 input and $1.5 output per 1M tokens; Kimi K3 at $0.192 and $9.5. DeepSeek V4 Pro is 4.3x cheaper per token at list price.
Which has the larger context window?
DeepSeek V4 Pro accepts 1M tokens and Kimi K3 1.0M tokens.