Compare / head-to-head
Grok 4vs
Kimi K2 Thinking
Grok 4 leads 3 of 3 shared benchmarks. Both offer a 256K-token context window.
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
3 – 0
Grok 4 leads
Cheaper per token
—
list price, input + output
Larger context
Tie
both 256K tokens
Providers
2 providers
xAI · Moonshot AI
Grok 4
xAI · released 2025-07-09
textvision
- Context
- 256K
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 17 · 8 core
Kimi K2 Thinking
Moonshot AI · released 2025-11-06
text
- Context
- 256K
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 17 · 9 core
Quality
Benchmark matrix
3 shared · 14 only Grok 4 · 14 only Kimi K2 Thinking| Benchmark | Grok 4 | Kimi K2 Thinking | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 3 | ||||
| GPQA (no tools) | 87.5% ↗ | 84.5% ↗ | +3 pt | Grok 4 |
| SimpleBench (SimpleBench) | 60.5% ↗ | 39.6% ↗ | +20.9 pt | Grok 4 |
| WeirdML (Håvard Tveit Ihle) | 45.7% ↗ | 42.8% ↗ | +2.9 pt | Grok 4 |
| Only Grok 4 reports · 14 | ||||
| GPQA Diamond | 87% ↗epoch run | not reported | — | — |
| AIME 2025 | 91.7% ↗ | not reported | — | — |
| Humanity's Last Exam (no tools) | 25.4% ↗ | not reported | — | — |
| LMArena Elo | 1410.8 ↗ | not reported | — | — |
| ARC-AGI-2 | 15.9% ↗ | not reported | — | — |
| BALROG (BALROG) | 43.6% ↗ | not reported | — | — |
| Chess Puzzles (Epoch AI run) | 28% ↗epoch run | not reported | — | — |
| HMMT 2025 (no tools) | 90% ↗ | not reported | — | — |
| Humanity's Last Exam (with tools) | 38.6% ↗ | not reported | — | — |
| LiveCodeBench (Jan - May) | 79% ↗ | not reported | — | — |
| LMArena Vision (LMArena) | 1184.2 ↗ | not reported | — | — |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 84% ↗epoch run | not reported | — | — |
| SAGE (Vals AI) | 25.1% ↗ | not reported | — | — |
| USAMO 2025 (no tools) | 37.5% ↗ | not reported | — | — |
| Only Kimi K2 Thinking reports · 14 | ||||
| AIME25 (no tools) | not reported | 94.5% ↗ | — | — |
| BrowseComp (w/ tools) | not reported | 60.2% ↗ | — | — |
| BrowseComp-ZH (w/ tools) | not reported | 62.3% ↗ | — | — |
| HLE (Text-only) (no tools) | not reported | 23.9% ↗ | — | — |
| HLE (Text-only) (w/ tools) | not reported | 44.9% ↗ | — | — |
| HMMT25 (no tools) | not reported | 89.4% ↗ | — | — |
| IMO-AnswerBench (no tools) | not reported | 78.6% ↗ | — | — |
| LiveCodeBenchV6 (no tools) | not reported | 83.1% ↗ | — | — |
| MMLU-Pro (no tools) | not reported | 84.6% ↗ | — | — |
| Multi-SWE-bench (w/ tools) | not reported | 41.9% ↗ | — | — |
| SciCode (no tools) | not reported | 44.8% ↗ | — | — |
| SWE-bench Multilingual (w/ tools) | not reported | 61.1% ↗ | — | — |
| SWE-bench Verified (w/ tools) | not reported | 71.3% ↗ | — | — |
| Terminal-Bench (w/ simulated tools (JSON)) | not reported | 47.1% ↗ | — | — |
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 Grok 4 minus Kimi K2 Thinking 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 | Grok 4 | Kimi K2 Thinking | Edge |
|---|---|---|---|
| Context window | 256K tokens | 256K tokens | Tie |
| Max output | — | — | — |
| Input price / 1M | — | — | — |
| Output price / 1M | — | — | — |
| Cached input / 1M | — | — | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | — | — | — |
| Modalities | text · vision | text | Grok 4 |
| Released | 2025-07-09 | 2025-11-06 | — |
| Cited benchmark scores | 17 | 17 | — |
Reliability
Provider status
Live from /statusMore matchups
Grok 4 vs …
Models sharing the most benchmarksMore matchups
Kimi K2 Thinking vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Grok 4 and Kimi K2 Thinking on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.