Qwen 3.8 Flash-Next
internal use only
Runs on our Sparks but is licensed for our own use only.
78
Logic Index · budget 4,096
53 t/s
tokens per second here
–
no tier here
Fact sheet
- Size
- NVFP4 (ModelOpt, MIXED_PRECISION)
- Context
- 256K
- Input
- Text, Image
- Capabilities
- ToolsImageReasoning
- License
- Qwen Community 1.0 +
NVIDIA Open Model License
Call
Internal: runs on our Sparks but is licensed for our own use only.
Logic Index
All models in the Logic Index →By test area
Reasoning chains
82
Logic vs. world knowledge
78
Logic grid puzzles
73
Consistency: same answer in every run for 76% of tasks.
Logically flawless in the part “Logic vs. world knowledge”; every deduction there is a misspelled answer (“UNGLÜLTIG” instead of “UNGÜLTIG”, German for “invalid”; the test is run in German). Scored as an error, since automated parsing fails on it just the same.
Index by thinking budget
Speed over time
Operations of the Sparks →Tokens per second from our nightly measurement. Latest: 53 t/s.