DeepSeek V4 Flash (max) vs MiniMax M3 (Reasoning)
Results updated 1 November 2026 · In Terminus-2 and MiniMax Code, so part of any gap can come from the harness · How we benchmark
Verdict
DeepSeek V4 Flash (max) edges MiniMax M3 (Reasoning) by 1 point, and is cheaper; MiniMax M3 (Reasoning) is faster.
- Cost per task
- $0.062 vs $0.15
- Time per task
- 5.7 min vs 4.9 min (estimated)
- Tokens per attempt
- 303k vs 1.68M
- Image input
- No vs Yes
DeepSeek V4 Flash (max), Terminus-2 scores higher on GeoBenchX and GS-QA2; MiniMax M3 (Reasoning), MiniMax Code scores higher on Earth-Bench V2. DeepSeek V4 Flash (max), Terminus-2 is cheaper ($0.062 against $0.15 per task), but MiniMax M3 (Reasoning), MiniMax Code is faster (4.9 min against 5.7 min).
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | DeepSeek V4 Flash (max) | MiniMax M3 (Reasoning) |
|---|---|---|
| Spatial statistics | 23 | 62 |
| Recognising infeasible requests | 53 | 33 |
| Climate and time series | 66 | 74 |
| Remote sensing and raster analysis | 78 | 73 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: DeepSeek V4 Flash (max), Terminus-2 · MiniMax M3 (Reasoning), MiniMax Code.
Side by side
| Metric | DeepSeek V4 Flash (max), Terminus-2 | MiniMax M3 (Reasoning), MiniMax Code |
|---|---|---|
| Geospatial Agent Index: Text only | 69 | 68 |
| Geospatial Agent Index: Text + vision | No image input | No data |
| Earth-Bench V2 score | 66 | 68 |
| GeoBenchX score | 62 | 59 |
| GS-QA2 score | 80 | 78 |
| Vision agents score | No image input | No data |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 96 | 94 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 95 | 79 |
| OGD4All score (standalone; in no index) | No data | 94 |
| GeoNatureAgent score (standalone; in no index) | No data | 93 |
| MultiGlobeQA score (standalone; in no index) | No data | 73 |
| GeoAgentEval score (standalone; in no index) | No data | 92 |
| Remote sensing and raster analysis score | 78 | 73 |
| GIS analysis score | 75 | 81 |
| Climate and time series score | 66 | 74 |
| Spatial statistics score | 23 | 62 |
| Making maps score | 79 | 77 |
| Recognising infeasible requests score | 53 | 33 |
| Cost per task | $0.062 | $0.15 |
| Time per task (at 100 output tokens/s) | 5.7 min | 4.9 min (estimated) |
| Tokens per attempt | 303k | 1.68M |
| Turns per attempt | 14.4 | No data |
| Input price per 1M tokens | $0.44 | $0.30 |
| Output price per 1M tokens | $1.32 | $1.20 |
| Context window | 1,048,576 | 1,048,576 |
| Image input | No | Yes |
| Developer | DeepSeek | MiniMax |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, DeepSeek V4 Flash (max), Terminus-2 or MiniMax M3 (Reasoning), MiniMax Code?
DeepSeek V4 Flash (max), Terminus-2, 69 to 68.
Which is cheaper per task, DeepSeek V4 Flash (max), Terminus-2 or MiniMax M3 (Reasoning), MiniMax Code?
DeepSeek V4 Flash (max), Terminus-2, $0.062 to $0.15.
Which writes fewer output tokens per task, DeepSeek V4 Flash (max), Terminus-2 or MiniMax M3 (Reasoning), MiniMax Code?
DeepSeek V4 Flash (max), Terminus-2, 29k to 29k.
Which has the larger context window, DeepSeek V4 Flash (max), Terminus-2 or MiniMax M3 (Reasoning), MiniMax Code?
They are level, at 1,048,576 tokens.