DeepSeek V4 Flash (high) vs Qwen 3.8 27B (Reasoning)
Results updated 24 October 2026 · In Codex and Qwen Code, so part of any gap can come from the harness · How we benchmark
Verdict
Qwen 3.8 27B (Reasoning) beats DeepSeek V4 Flash (high) by 3 points, and is faster; DeepSeek V4 Flash (high) is cheaper.
- Cost per task
- $0.085 vs $0.22
- Time per task
- 6.5 min vs 4.7 min (estimated)
- Tokens per attempt
- 1.13M vs 840k
- Image input
- No vs Yes
DeepSeek V4 Flash (high), Codex scores higher on Earth-Bench V2; Qwen 3.8 27B (Reasoning), Qwen Code scores higher on GeoBenchX and GS-QA2. DeepSeek V4 Flash (high), Codex is cheaper ($0.085 against $0.22 per task), but Qwen 3.8 27B (Reasoning), Qwen Code is faster (4.7 min against 6.5 min).
Where they differ most
By kind of work, share of tasks passed, 0 to 100 · Higher is better
| Kind of work | DeepSeek V4 Flash (high) | Qwen 3.8 27B (Reasoning) |
|---|---|---|
| Recognising infeasible requests | 37 | 62 |
| Spatial statistics | 42 | 54 |
| Climate and time series | 73 | 63 |
| Making maps | 70 | 65 |
Where the tasks are set: each kind of work links to its tasks and scoresCharts for each: DeepSeek V4 Flash (high), Codex · Qwen 3.8 27B (Reasoning), Qwen Code.
Side by side
| Metric | DeepSeek V4 Flash (high), Codex | Qwen 3.8 27B (Reasoning), Qwen Code |
|---|---|---|
| Geospatial Agent Index: Text only | 66 | 69 |
| Geospatial Agent Index: Text + vision | No image input | No data |
| Earth-Bench V2 score | 70 | 69 |
| GeoBenchX score | 54 | 63 |
| GS-QA2 score | 74 | 75 |
| Vision agents score | No image input | No data |
| GeoAgentBench score (Guided workflows; not in the Agent Index) | 92 | 94 |
| GeoAnalystBench score (Guided workflows; not in the Agent Index) | 89 | 79 |
| OGD4All score (standalone; in no index) | No data | 97 |
| GeoNatureAgent score (standalone; in no index) | No data | 100 |
| MultiGlobeQA score (standalone; in no index) | No data | 74 |
| GeoAgentEval score (standalone; in no index) | No data | 88 |
| ImpactMesh score (standalone; in no index) | No data | No data |
| Remote sensing and raster analysis score | 72 | 77 |
| GIS analysis score | 75 | 75 |
| Climate and time series score | 73 | 63 |
| Spatial statistics score | 42 | 54 |
| Making maps score | 70 | 65 |
| Recognising infeasible requests score | 37 | 62 |
| Cost per task | $0.085 | $0.22 |
| Time per task (at 100 output tokens/s) | 6.5 min | 4.7 min (estimated) |
| Tokens per attempt | 1.13M | 840k |
| Turns per attempt | 25.4 | 17.6 |
| Input price per 1M tokens | $0.44 | $0.45 |
| Output price per 1M tokens | $1.32 | $3.20 |
| Context window | 1,048,576 | 262,144 |
| Image input | No | Yes |
| Developer | DeepSeek | Alibaba (Qwen) |
Frequently asked questions
Which scores higher on the Geospatial Agent Index, DeepSeek V4 Flash (high), Codex or Qwen 3.8 27B (Reasoning), Qwen Code?
Qwen 3.8 27B (Reasoning), Qwen Code, 69 to 66.
Which is cheaper per task, DeepSeek V4 Flash (high), Codex or Qwen 3.8 27B (Reasoning), Qwen Code?
DeepSeek V4 Flash (high), Codex, $0.085 to $0.22.
Which writes fewer output tokens per task, DeepSeek V4 Flash (high), Codex or Qwen 3.8 27B (Reasoning), Qwen Code?
DeepSeek V4 Flash (high), Codex, 19k to 28k.
Which has the larger context window, DeepSeek V4 Flash (high), Codex or Qwen 3.8 27B (Reasoning), Qwen Code?
DeepSeek V4 Flash (high), Codex, 1,048,576 tokens to 262,144 tokens.