Kimi K3 (Reasoning) vs GPT-6.1 Sol (medium)
Results updated 31 October 2026 · In Kimi Code and Codex, so part of any gap can come from the harness · How we benchmark
Verdict
GPT-6.1 Sol (medium) beats Kimi K3 (Reasoning) by 7 points, and is cheaper and faster.
- Cost per task
- $0.32 vs $0.046
- Time per task
- 1.5 min (estimated) vs 0.4 min
- Tokens per attempt
- 345k vs 89k
- Image input
- Yes vs Yes
Kimi K3 (Reasoning), Kimi Code scores higher on Earth-Bench V2; GPT-6.1 Sol (medium), Codex scores higher on GeoBenchX and GS-QA2. GPT-6.1 Sol (medium), Codex is cheaper and faster: $0.046 against $0.32 per task, and 0.4 min against 1.5 min.
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | Kimi K3 (Reasoning) | GPT-6.1 Sol (medium) |
|---|---|---|
| Recognising infeasible requests | 52 | 90 |
| GIS analysis | 84 | 94 |
| Spatial statistics | 38 | 31 |
| Climate and time series | 78 | 75 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: Kimi K3 (Reasoning), Kimi Code · GPT-6.1 Sol (medium), Codex.
Side by side
| Metric | Kimi K3 (Reasoning), Kimi Code | GPT-6.1 Sol (medium), Codex |
|---|---|---|
| Geospatial Agent Index: Text only | 75 | 82 |
| Geospatial Agent Index: Text + vision | No data | No data |
| Earth-Bench V2 score | 74 | 72 |
| GeoBenchX score | 65 | 79 |
| GS-QA2 score | 85 | 93 |
| Vision agents score | No data | No data |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 96 | 100 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 100 | 100 |
| OGD4All score (standalone; in no index) | 97 | No data |
| GeoNatureAgent score (standalone; in no index) | 100 | No data |
| MultiGlobeQA score (standalone; in no index) | 77 | No data |
| GeoAgentEval score (standalone; in no index) | 96 | No data |
| Remote sensing and raster analysis score | 81 | 83 |
| GIS analysis score | 84 | 94 |
| Climate and time series score | 78 | 75 |
| Spatial statistics score | 38 | 31 |
| Making maps score | 82 | 81 |
| Recognising infeasible requests score | 52 | 90 |
| Cost per task | $0.32 | $0.046 |
| Time per task (at 100 output tokens/s) | 1.5 min (estimated) | 0.4 min |
| Tokens per attempt | 345k | 89k |
| Turns per attempt | No data | 5.8 |
| Input price per 1M tokens | $3.00 | $2.00 |
| Output price per 1M tokens | $15.00 | $10.00 |
| Context window | 1,000,000 | 1,050,000 |
| Image input | Yes | Yes |
| Developer | Moonshot AI | OpenAI |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, Kimi K3 (Reasoning), Kimi Code or GPT-6.1 Sol (medium), Codex?
GPT-6.1 Sol (medium), Codex, 82 to 75.
Which is cheaper per task, Kimi K3 (Reasoning), Kimi Code or GPT-6.1 Sol (medium), Codex?
GPT-6.1 Sol (medium), Codex, $0.046 to $0.32.
Which writes fewer output tokens per task, Kimi K3 (Reasoning), Kimi Code or GPT-6.1 Sol (medium), Codex?
GPT-6.1 Sol (medium), Codex, 1k to 9k.
Which has the larger context window, Kimi K3 (Reasoning), Kimi Code or GPT-6.1 Sol (medium), Codex?
GPT-6.1 Sol (medium), Codex, 1,050,000 tokens to 1,000,000 tokens.