Claude Haiku 5.5 (max) vs GPT-6.1 Sol (medium)
Results updated 4 November 2026 · In Claude Code and Codex, so part of any gap can come from the harness · How we benchmark
Verdict
GPT-6.1 Sol (medium) beats Claude Haiku 5.5 (max) by 5 points, and is cheaper and faster.
- Cost per task
- $0.087 vs $0.046
- Time per task
- 8.2 min (estimated) vs 0.4 min
- Tokens per attempt
- 930k vs 89k
- Image input
- Yes vs Yes
GPT-6.1 Sol (medium), Codex scores higher than Claude Haiku 5.5 (max), Claude Code on all 3 evaluations scored for both. GPT-6.1 Sol (medium), Codex is cheaper and faster: $0.046 against $0.087 per task, and 0.4 min against 8.2 min.
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | Claude Haiku 5.5 (max) | GPT-6.1 Sol (medium) |
|---|---|---|
| Spatial statistics | 46 | 31 |
| Recognising infeasible requests | 83 | 90 |
| Remote sensing and raster analysis | 77 | 83 |
| GIS analysis | 90 | 94 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: Claude Haiku 5.5 (max), Claude Code · GPT-6.1 Sol (medium), Codex.
Side by side
| Metric | Claude Haiku 5.5 (max), Claude Code | GPT-6.1 Sol (medium), Codex |
|---|---|---|
| Geospatial Agent Index: Text only | 77 | 82 |
| Geospatial Agent Index: Text + vision | 78 | No data |
| Earth-Bench V2 score | 67 | 72 |
| GeoBenchX score | 77 | 79 |
| GS-QA2 score | 87 | 93 |
| Vision agents score | 81 | No data |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 98 | 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) | 76 | No data |
| GeoAgentEval score (standalone; in no index) | 96 | No data |
| Remote sensing and raster analysis score | 77 | 83 |
| GIS analysis score | 90 | 94 |
| Climate and time series score | 71 | 75 |
| Spatial statistics score | 46 | 31 |
| Making maps score | 83 | 81 |
| Recognising infeasible requests score | 83 | 90 |
| Cost per task | $0.087 | $0.046 |
| Time per task (at 100 output tokens/s) | 8.2 min (estimated) | 0.4 min |
| Tokens per attempt | 930k | 89k |
| Turns per attempt | 13.6 | 5.8 |
| Input price per 1M tokens | $0.10 | $2.00 |
| Output price per 1M tokens | $0.50 | $10.00 |
| Context window | 1,000,000 | 1,050,000 |
| Image input | Yes | Yes |
| Developer | Anthropic | OpenAI |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, Claude Haiku 5.5 (max), Claude Code or GPT-6.1 Sol (medium), Codex?
GPT-6.1 Sol (medium), Codex, 82 to 77.
Which is cheaper per task, Claude Haiku 5.5 (max), Claude Code or GPT-6.1 Sol (medium), Codex?
GPT-6.1 Sol (medium), Codex, $0.046 to $0.087.
Which writes fewer output tokens per task, Claude Haiku 5.5 (max), Claude Code or GPT-6.1 Sol (medium), Codex?
GPT-6.1 Sol (medium), Codex, 1k to 45k.
Which has the larger context window, Claude Haiku 5.5 (max), Claude Code or GPT-6.1 Sol (medium), Codex?
GPT-6.1 Sol (medium), Codex, 1,050,000 tokens to 1,000,000 tokens.