XOGENT
Runs814bd618

deepseek/deepseek-v4.1-flash

publishedsuite v1.0opencode opencode 1.18.30 · started Sep 21, 2026, 08:47 PM · finished Sep 21, 2026, 08:54 PM

Models
1
Tasks attempted
65
Attempts
195
3 repeats per task per model
Evaluations
1436
deterministic + judge + human

Published as a provisional synthetic-fixture panel assessment. Pending review flags were accepted with the documented coverage limits; attempts were not individually rescored.

Models in this run

Pricing snapshot captured when the run started, per 1M tokens.

ModelEffortInputOutputMean score
DeepSeek V4.1 Flashdeepseek/deepseek-v4.1-flash
max––63.0

Anchor drift

Per anchor model, category score in this run minus the previous published run. More than 5 points in either direction holds the run.

No drift check recorded

This run had no previous published run to compare against, or the check was not recorded.

Category leaders in this run

3D and Graphics
  1. 1DeepSeek V4.1 Flash0.6
Coding
  1. 1DeepSeek V4.1 Flash57.2
Data and SQL
  1. 1DeepSeek V4.1 Flash56.5
Debugging
  1. 1DeepSeek V4.1 Flash56.4
IT Scripting
  1. 1DeepSeek V4.1 Flash0.0
Security
  1. 1DeepSeek V4.1 Flash72.3
UI Design
  1. 1DeepSeek V4.1 Flash28.8
Client Communication
  1. 1DeepSeek V4.1 Flash93.6
Document Extraction
  1. 1DeepSeek V4.1 Flash92.0
Hallucination Resistance
  1. 1DeepSeek V4.1 Flash94.7
Legal Drafting
  1. 1DeepSeek V4.1 Flash93.4
Reasoning and Analysis
  1. 1DeepSeek V4.1 Flash77.8
Summarization
  1. 1DeepSeek V4.1 Flash95.5