Claude Haiku 4.5 (Reasoning), Claude Code vs. Qwen 3.8 27B (xhigh), Qwen Code
Claude Haiku 4.5 (Reasoning), Claude Code against Qwen 3.8 27B (xhigh), Qwen Code: Geospatial Agent Index, each evaluation, cost, time and tokens, on the same tasks and limits. The harnesses differ, so part of any gap can come from the harness. Methodology.
Side by side
| Metric | Claude Haiku 4.5 (Reasoning), Claude Code | Qwen 3.8 27B (xhigh), Qwen Code |
|---|---|---|
| Geospatial Agent Index: Text only | 58 | 68 |
| Geospatial Agent Index: Text + vision | 48 | Not run: Vision agents |
| Earth-Bench V2 score | 51 | 68 |
| GeoBenchX score | 53 | 62 |
| GS-QA2 score | 69 | 75 |
| Vision agents score | 19 | No data |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 78 | 94 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 74 | 79 |
| Raster and remote sensing score | 65 | 76 |
| Vector and overlay score | 67 | 75 |
| Climate and time series score | 49 | 61 |
| Spatial statistics and interpolation score | 23 | 54 |
| Mapping and cartography score | 56 | 63 |
| Recognising infeasible requests score | 53 | 60 |
| Cost per task | $0.092 | $0.22 |
| Time per task (at 100 output tokens/s) | 1.6 min | 4.7 min (estimated) |
| Tokens per attempt | 313k | 840k |
| Turns per attempt | 10.6 | 17.6 |
| Attempts scored | 488 | 488 |
| Input price per 1M tokens | $1.00 | $0.45 |
| Output price per 1M tokens | $5.00 | $3.20 |
| Context window | 200,000 | 262,144 |
| Image input | Yes | Yes |
| Developer | Anthropic | Alibaba (Qwen) |
Charts
Geospatial Agent Index
Equal average of Earth-Bench V2, GeoBenchX and GS-QA2, 0 to 100 · Higher is better
Top eight. Every model is in the table.
Data table: Geospatial Agent Index
| Model | Developer | Geospatial Agent Index | Status | Evaluations scored | Tasks attempted |
|---|---|---|---|---|---|
| Qwen 3.8 27B (xhigh), Qwen Code | Alibaba (Qwen) | 68 | Complete | 3 of 3 | 488 of 488 tasks |
| Claude Haiku 4.5 (Reasoning), Claude Code | Anthropic | 58 | Complete | 3 of 3 | 488 of 488 tasks |
Cost per task
Average cost per task (USD) · Lower is better
Top eight. Every model is in the table.
Data table: Cost per task
| Model | Developer | Cost per task | Tasks attempted |
|---|---|---|---|
| Claude Haiku 4.5 (Reasoning), Claude Code | Anthropic | $0.092 | 488 of 488 tasks |
| Qwen 3.8 27B (xhigh), Qwen Code | Alibaba (Qwen) | $0.22 | 488 of 488 tasks |
Output tokens per task
Average output tokens per attempt, reasoning included · Lower is better
Top eight. Every model is in the table.
Data table: Output tokens per task
| Model | Developer | Output tokens per task | Tasks attempted |
|---|---|---|---|
| Claude Haiku 4.5 (Reasoning), Claude Code | Anthropic | 7k | 488 of 488 tasks |
| Qwen 3.8 27B (xhigh), Qwen Code | Alibaba (Qwen) | 28k | 488 of 488 tasks |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, Claude Haiku 4.5 (Reasoning), Claude Code or Qwen 3.8 27B (xhigh), Qwen Code?
Qwen 3.8 27B (xhigh), Qwen Code, 68 to 58. Both have attempted all 488 tasks.
Which is cheaper per task, Claude Haiku 4.5 (Reasoning), Claude Code or Qwen 3.8 27B (xhigh), Qwen Code?
Claude Haiku 4.5 (Reasoning), Claude Code, $0.092 to $0.22.
Which writes fewer output tokens per task, Claude Haiku 4.5 (Reasoning), Claude Code or Qwen 3.8 27B (xhigh), Qwen Code?
Claude Haiku 4.5 (Reasoning), Claude Code, 7k to 28k.
Which has the larger context window, Claude Haiku 4.5 (Reasoning), Claude Code or Qwen 3.8 27B (xhigh), Qwen Code?
Qwen 3.8 27B (xhigh), Qwen Code, 262,144 tokens to 200,000 tokens.
More: Claude Haiku 4.5 (Reasoning), Claude Code · Qwen 3.8 27B (xhigh), Qwen Code · All comparisons