MiniMax M3 (Reasoning) vs Claude Haiku 5.5 (low)
Results updated 1 November 2026 · In MiniMax Code and Claude Code, so part of any gap can come from the harness · How we benchmark
Verdict
Claude Haiku 5.5 (low) edges MiniMax M3 (Reasoning) by 2 points, and is cheaper and faster.
- Cost per task
- $0.15 vs $0.0053
- Time per task
- 4.9 min (estimated) vs 0.9 min
- Tokens per attempt
- 1.68M vs 181k
- Image input
- Yes vs Yes
MiniMax M3 (Reasoning), MiniMax Code scores higher on Earth-Bench V2 and GS-QA2; Claude Haiku 5.5 (low), Claude Code scores higher on GeoBenchX. Claude Haiku 5.5 (low), Claude Code is cheaper and faster: $0.0053 against $0.15 per task, and 0.9 min against 4.9 min.
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | MiniMax M3 (Reasoning) | Claude Haiku 5.5 (low) |
|---|---|---|
| Recognising infeasible requests | 33 | 73 |
| Spatial statistics | 62 | 38 |
| Climate and time series | 74 | 64 |
| GIS analysis | 81 | 73 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: MiniMax M3 (Reasoning), MiniMax Code · Claude Haiku 5.5 (low), Claude Code.
Side by side
| Metric | MiniMax M3 (Reasoning), MiniMax Code | Claude Haiku 5.5 (low), Claude Code |
|---|---|---|
| Geospatial Agent Index: Text only | 68 | 70 |
| Geospatial Agent Index: Text + vision | No data | 72 |
| Earth-Bench V2 score | 68 | 62 |
| GeoBenchX score | 59 | 73 |
| GS-QA2 score | 78 | 75 |
| Vision agents score | No data | 77 |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 94 | 92 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 79 | 100 |
| OGD4All score (standalone; in no index) | 94 | 95 |
| GeoNatureAgent score (standalone; in no index) | 93 | 100 |
| MultiGlobeQA score (standalone; in no index) | 73 | 72 |
| GeoAgentEval score (standalone; in no index) | 92 | 92 |
| Remote sensing and raster analysis score | 73 | 72 |
| GIS analysis score | 81 | 73 |
| Climate and time series score | 74 | 64 |
| Spatial statistics score | 62 | 38 |
| Making maps score | 77 | 79 |
| Recognising infeasible requests score | 33 | 73 |
| Cost per task | $0.15 | $0.0053 |
| Time per task (at 100 output tokens/s) | 4.9 min (estimated) | 0.9 min |
| Tokens per attempt | 1.68M | 181k |
| Turns per attempt | No data | 5.5 |
| Input price per 1M tokens | $0.30 | $0.10 |
| Output price per 1M tokens | $1.20 | $0.50 |
| Context window | 1,048,576 | 1,000,000 |
| Image input | Yes | Yes |
| Developer | MiniMax | Anthropic |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, MiniMax M3 (Reasoning), MiniMax Code or Claude Haiku 5.5 (low), Claude Code?
Claude Haiku 5.5 (low), Claude Code, 70 to 68.
Which is cheaper per task, MiniMax M3 (Reasoning), MiniMax Code or Claude Haiku 5.5 (low), Claude Code?
Claude Haiku 5.5 (low), Claude Code, $0.0053 to $0.15.
Which writes fewer output tokens per task, MiniMax M3 (Reasoning), MiniMax Code or Claude Haiku 5.5 (low), Claude Code?
Claude Haiku 5.5 (low), Claude Code, 3k to 29k.
Which has the larger context window, MiniMax M3 (Reasoning), MiniMax Code or Claude Haiku 5.5 (low), Claude Code?
MiniMax M3 (Reasoning), MiniMax Code, 1,048,576 tokens to 1,000,000 tokens.