Claude Sonnet 5.5 (high) vs Claude Haiku 5.5 (max)
Results updated 3 November 2026 · Both in Claude Code · How we benchmark
Verdict
Claude Sonnet 5.5 (high) and Claude Haiku 5.5 (max) are level on the Geospatial Agent Index, at 78.
- Cost per task
- $0.055 vs $0.079
- Time per task
- 0.6 min vs 7.8 min (estimated)
- Tokens per attempt
- 94k vs 859k
- Image input
- Yes vs Yes
Claude Sonnet 5.5 (high), Claude Code scores higher on Earth-Bench V2 and GeoBenchX; Claude Haiku 5.5 (max), Claude Code scores higher on GS-QA2 and Vision agents. Claude Sonnet 5.5 (high), Claude Code is cheaper and faster: $0.055 against $0.079 per task, and 0.6 min against 7.8 min.
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | Claude Sonnet 5.5 (high) | Claude Haiku 5.5 (max) |
|---|---|---|
| Spatial statistics | 62 | 45 |
| Climate and time series | 69 | 76 |
| Recognising infeasible requests | 80 | 86 |
| GIS analysis | 86 | 91 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: Claude Sonnet 5.5 (high), Claude Code · Claude Haiku 5.5 (max), Claude Code.
Side by side
| Metric | Claude Sonnet 5.5 (high), Claude Code | Claude Haiku 5.5 (max), Claude Code |
|---|---|---|
| Geospatial Agent Index: Text only | 78 | 78 |
| Geospatial Agent Index: Text + vision | 78 | 79 |
| Earth-Bench V2 score | 71 | 70 |
| GeoBenchX score | 81 | 78 |
| GS-QA2 score | 83 | 87 |
| Vision agents score | 78 | 81 |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 98 | 97 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 100 | 100 |
| OGD4All score (standalone; in no index) | 96 | 97 |
| GeoNatureAgent score (standalone; in no index) | 100 | 100 |
| MultiGlobeQA score (standalone; in no index) | 77 | 76 |
| GeoAgentEval score (standalone; in no index) | 100 | 96 |
| Remote sensing and raster analysis score | 80 | 77 |
| GIS analysis score | 86 | 91 |
| Climate and time series score | 69 | 76 |
| Spatial statistics score | 62 | 45 |
| Making maps score | 84 | 85 |
| Recognising infeasible requests score | 80 | 86 |
| Cost per task | $0.055 | $0.079 |
| Time per task (at 100 output tokens/s) | 0.6 min | 7.8 min (estimated) |
| Tokens per attempt | 94k | 859k |
| Turns per attempt | 4.8 | 13.1 |
| Input price per 1M tokens | $2.00 | $0.10 |
| Output price per 1M tokens | $10.00 | $0.50 |
| Context window | 1,000,000 | 1,000,000 |
| Image input | Yes | Yes |
| Developer | Anthropic | Anthropic |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, Claude Sonnet 5.5 (high), Claude Code or Claude Haiku 5.5 (max), Claude Code?
Claude Sonnet 5.5 (high), Claude Code, 78 to 78.
Which is cheaper per task, Claude Sonnet 5.5 (high), Claude Code or Claude Haiku 5.5 (max), Claude Code?
Claude Sonnet 5.5 (high), Claude Code, $0.055 to $0.079.
Which writes fewer output tokens per task, Claude Sonnet 5.5 (high), Claude Code or Claude Haiku 5.5 (max), Claude Code?
Claude Sonnet 5.5 (high), Claude Code, 2k to 43k.
Which has the larger context window, Claude Sonnet 5.5 (high), Claude Code or Claude Haiku 5.5 (max), Claude Code?
They are level, at 1,000,000 tokens.