Compare / head-to-head
Kimi K3vs
Mercury 2.5
Kimi K3 leads 1 of 1 shared benchmark. Mercury 2.5 is 18.9x cheaper per token. Kimi K3 has the larger context window (1.0M tokens).
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
1 – 0
Kimi K3 leads
Cheaper per token
Mercury 2.5
18.9x cheaper, input + output
Larger context
Kimi K3
1.0M tokens
Providers
2 providers
Moonshot AI · Inception
Kimi K3
Moonshot AI · released 2026-07-16
textvision
Mercury 2.5
Inception · released 2026-09-08
text
Quality
Benchmark matrix
1 shared · 79 only Kimi K3 · 0 only Mercury 2.5| Benchmark | Kimi K3 | Mercury 2.5 | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 1 | ||||
| Terminal-Bench 4.0 (Vals AI) | 17.2% ↗ | 0% ↗ | +17.2 pt | Kimi K3 |
| Only Kimi K3 reports · 79 | ||||
| GPQA Diamond | 93.5% ↗ | not reported | — | — |
| AIME 2026 | 96.7% ↗matharena ⚠ | not reported | — | — |
| Humanity's Last Exam (no tools) | 43.5% ↗ | not reported | — | — |
| LMArena Elo | 1488.4 ↗ | not reported | — | — |
| AA-Briefcase | 1548 ↗ | not reported | — | — |
| AA-LCR | 74.7% ↗ | not reported | — | — |
| Agents' Last Exam | 28.3% ↗ | not reported | — | — |
| APEX-Agents | 41% ↗ | not reported | — | — |
| APEX-Agents (Mercor) | 50.6% ↗ | not reported | — | — |
| AutomationBench | 30.8% ↗ | not reported | — | — |
| BabyVision (with Python) | 85.7% ↗ | not reported | — | — |
| BrowseComp (1M context, no compaction) | 90.4% ↗ | not reported | — | — |
| BrowseComp (context compaction) | 91.2% ↗ | not reported | — | — |
| CharXiv Reasoning (no tools) | 84.8% ↗ | not reported | — | — |
| CharXiv Reasoning (with Python) | 91.3% ↗ | not reported | — | — |
| Chess Puzzles (Epoch AI run) | 39% ↗epoch run | not reported | — | — |
| CorpFin v2 | 71.6% ↗ | not reported | — | — |
| CritPt | 23.4% ↗ | not reported | — | — |
| DeepSearchQA (F1) | 95% ↗ | not reported | — | — |
| DeepSWE v1.1 | 67.3% ↗ | not reported | — | — |
| DeepSWE v1.1 (Datacurve) | 68.5% ↗ | not reported | — | — |
| DeepSWE v1.1 (Kimi Code) | 67.5% ↗ | not reported | — | — |
| Finance Agent v2 | 54.4% ↗ | not reported | — | — |
| FrontierMath Tier 4 v2 (Epoch AI run) | 39% ↗epoch run | not reported | — | — |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 72.2% ↗epoch run | not reported | — | — |
| FrontierSWE | 81.2% ↗ | not reported | — | — |
| FrontierSWE V2 (Proximal Labs) | 25.9% ↗ | not reported | — | — |
| Furniture Assembly (Epoch AI run) | 34.2% ↗epoch run | not reported | — | — |
| GDPval-AA v2 | 1686 ↗ | not reported | — | — |
| Harvey Lab-AA | 94.6% ↗ | not reported | — | — |
| Humanity's Last Exam (with tools) | 56% ↗ | not reported | — | — |
| JobBench | 54.3% ↗ | not reported | — | — |
| Kimi Code Bench 2.0 | 72.9% ↗ | not reported | — | — |
| Legal Research Bench | 44.2% ↗ | not reported | — | — |
| LiveBench Agentic Coding (LiveBench) | 62.2% ↗ | not reported | — | — |
| LiveBench Coding (LiveBench) | 81.4% ↗ | not reported | — | — |
| LiveBench Data Analysis (LiveBench) | 78.7% ↗ | not reported | — | — |
| LiveBench Instruction Following (LiveBench) | 71.4% ↗ | not reported | — | — |
| LiveBench Language (LiveBench) | 85.5% ↗ | not reported | — | — |
| LiveBench Mathematics (LiveBench) | 84.4% ↗ | not reported | — | — |
| LiveBench Reasoning (LiveBench) | 90.7% ↗ | not reported | — | — |
| LMArena Agent (LMArena) | 0.0418 ↗ | not reported | — | — |
| LMArena WebDev (LMArena) | 1657.8 ↗ | not reported | — | — |
| MathVision (no tools) | 94.3% ↗ | not reported | — | — |
| MathVision (with Python) | 97.8% ↗ | not reported | — | — |
| MCP-Atlas | 84.2% ↗ | not reported | — | — |
| MCPMark-Verified | 94.5% ↗ | not reported | — | — |
| MLS-Bench-Lite | 48.3% ↗ | not reported | — | — |
| MMMU-Pro (no tools) | 81.6% ↗ | not reported | — | — |
| MMMU-Pro (with Python) | 83.4% ↗ | not reported | — | — |
| MMVU | 82.1% ↗ | not reported | — | — |
| Mystery Game Puzzles (Epoch AI run) | 26% ↗epoch run | not reported | — | — |
| OfficeQA Pro | 63.3% ↗ | not reported | — | — |
| OmniDocBench | 91.1% ↗ | not reported | — | — |
| OSWorld 2.0 | 58.3% ↗ | not reported | — | — |
| OSWorld-Verified | 84.8% ↗ | not reported | — | — |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 97.2% ↗epoch run | not reported | — | — |
| PerceptionBench | 58.5% ↗ | not reported | — | — |
| PostTrainBench | 36.6% ↗ | not reported | — | — |
| ProgramBench | 77.8% ↗ | not reported | — | — |
| ResearchRubrics | 76.2% ↗ | not reported | — | — |
| SaaS-Bench | 60.1% ↗ | not reported | — | — |
| SAGE (Vals AI) | 52.8% ↗ | not reported | — | — |
| SciCode | 58.7% ↗ | not reported | — | — |
| SimpleBench (SimpleBench) | 60.7% ↗ | not reported | — | — |
| SimpleQA Verified | 50.6% ↗epoch run | not reported | — | — |
| SpreadsheetBench 2 | 34.8% ↗ | not reported | — | — |
| SWE-Marathon | 42% ↗ | not reported | — | — |
| tau2-bench Banking Knowledge (Sierra) | 37.1% ↗ | not reported | — | — |
| tau3-Banking | 33.4% ↗ | not reported | — | — |
| Terminal-Bench 2.1 | 88.3% ↗ | not reported | — | — |
| Toolathlon-Verified | 76.5% ↗ | not reported | — | — |
| Toolathlon-Verified (HKUST) | 76.5% ↗ | not reported | — | — |
| Vending-Bench 2 (Andon Labs) | 5165.04 ↗ | not reported | — | — |
| Video-MME (with subtitles) | 90% ↗ | not reported | — | — |
| WeirdML (Håvard Tveit Ihle) | 82.6% ↗ | not reported | — | — |
| WorldVQA ForceAnswer | 51% ↗ | not reported | — | — |
| ZeroBench pass@5 (no tools) | 23% ↗ | not reported | — | — |
| ZeroBench pass@5 (with Python) | 41% ↗ | not reported | — | — |
| Only Mercury 2.5 reports · 0 | ||||
| Mercury 2.5 reports nothing Kimi K3 does not. | ||||
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 Kimi K3 minus Mercury 2.5 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 | Kimi K3 | Mercury 2.5 | Edge |
|---|---|---|---|
| Context window | 1.0M tokens | 260K tokens | Kimi K3 |
| Max output | 1.0M tokens | 66K tokens | Kimi K3 |
| Input price / 1M | $3 ↗ | $0.2 ↗ | Mercury 2.5 |
| Output price / 1M | $15 ↗ | $0.75 ↗ | Mercury 2.5 |
| Cached input / 1M | $0.3 ↗ | $0.02 ↗ | Mercury 2.5 |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $18 | $0.95 | Mercury 2.5 |
| Modalities | text · vision | text | Kimi K3 |
| Released | 2026-07-16 | 2026-09-08 | — |
| Cited benchmark scores | 87 | 1 | — |
Reliability
Provider status
Live from /statusMore matchups
Kimi K3 vs …
Models sharing the most benchmarksMore matchups
Mercury 2.5 vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Kimi K3 and Mercury 2.5 on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.