Kaelum Bench v0.2 · Public Calibration

deepseek-v4-flash

HGB Reporting · Run evidence

Provisional · Deterministic

Provisional calibration result — not a verified v1 result.

Configuration-specific evidence for a single published run. It is not a statistical ranking and is not an overall score.

Run identity

Model and configuration

Model display name
deepseek-v4-flash
Requested model ID
deepseek/deepseek-v4-flash
Provider
Alibaba Cloud
Reasoning configuration
xhigh
Suite
HGB Reporting
Validity
Provisional

Published score

Scoring evidence

Score
507.1 / 600
Percentage
84.5%
Scoring method
Deterministic
Comparison protocol
Configuration-specific
Tool access
None
Published total cost
$0.0483

Usage

Tokens and execution

Input tokens (tested model)
68,464
Output tokens (tested model)
145,982
Total tokens (tested model)
214,446
Attempted tasks
60 / 60
Valid responses
60 / 60
Invalid JSON
0
Schema failures
0
Provider failures
0
Missing responses
0

Release provenance

Versioned run evidence

Benchmark ID
FMB-KNOW-ACC-HGB-02
Benchmark version
0.2.0-dev
Scorer ID
fmb-live-scorer-v1
Started at
Completed at
Run ID
b267168d-0c48-5501-904b-40b1a3d01e1a
Signed release ID
82d3a8e7-8cba-42c0-a30c-8cf7e38256aa