Claude Haiku 5.5 (max) vs GPT-6 Sol (medium)
Results updated 3 November 2026 · In Claude Code and Codex, so part of any gap can come from the harness · How we benchmark
Verdict
Not comparable yet: GPT-6 Sol (medium) is not yet ranked.
- Cost per task
- $0.079 vs $0.047
- Time per task
- 7.8 min (estimated) vs 0.3 min
- Tokens per attempt
- 859k vs 86k
- Image input
- Yes vs Yes
Claude Haiku 5.5 (max), Claude Code scores higher on GeoBenchX; GPT-6 Sol (medium), Codex scores higher on Earth-Bench V2 and Vision agents; they are level on GS-QA2. GPT-6 Sol (medium), Codex is cheaper and faster: $0.047 against $0.079 per task, and 0.3 min against 7.8 min.
Not yet ranked. GPT-6 Sol (medium), Codex. Not in the Agent Index, cost and time charts until it has a score on every index evaluation. Results so far are in the side-by-side table.
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 Sol (medium) |
|---|---|---|
| Making maps | 85 | 100 |
| Recognising infeasible requests | 86 | 75 |
| Remote sensing and raster analysis | 77 | 88 |
| GIS analysis | 91 | 83 |
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 Sol (medium), Codex.
Side by side
| Metric | Claude Haiku 5.5 (max), Claude Code | GPT-6 Sol (medium), Codex |
|---|---|---|
| Geospatial Agent Index: Text only | 78 | 80 |
| Geospatial Agent Index: Text + vision | 79 | No data |
| Earth-Bench V2 score | 70 | 77 |
| GeoBenchX score | 78 | 75 |
| GS-QA2 score | 87 | 87 |
| Vision agents score | 81 | 91 |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 97 | 100 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 100 | No data |
| 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 | 88 |
| GIS analysis score | 91 | 83 |
| Climate and time series score | 76 | 76 |
| Spatial statistics score | 45 | 50 |
| Making maps score | 85 | 100 |
| Recognising infeasible requests score | 86 | 75 |
| Cost per task | $0.079 | $0.047 |
| Time per task (at 100 output tokens/s) | 7.8 min (estimated) | 0.3 min |
| Tokens per attempt | 859k | 86k |
| Turns per attempt | 13.1 | 6.4 |
| 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 |