chat model, answer-letter probabilities · v1 set (earlier version)

DeepSeek-V4.1-Flash

Earlier version

Results f24d1f41, updated 16 November 2026, 22:20 UTC; scored by GNU Backgammon 1.08, 2-ply.

How we called it

Model ID
deepseek-v4.1-flash
Provider or endpoint
our GPUs (SGLang) behind Cygnet's letter-reading server
Date run
Missing
Price
no price per token; own hardware; price date not recorded
Model type
chat model, answer-letter probabilities
Weights
open
Runs on
our GPUs
Reasoning setting
Missing
Confidence source
letter probability
Rules text
short (the v1 rules text; the same stored request text for every model)
Option groups
not recorded; no decision in this data was split
Sampling settings
temperature 1.0
Notes
Missing

Missing B7 fields: reasoning, option_groups, date_run

Results in v1 set (earlier version)

Earlier version: plays were listed in the runner's order, which tends to put the best play early. When Gemini 4 Argon answered the same 1,000 positions with the plays shuffled, its error rate changed by +1.9 (95% interval −0.2 to +4.2), so the effect is not proven.

Error rate (PR)
27.523.7 to 31.3
Decisions
1,000 (1,000 scored, 0 invalid)
Engine's best play
51.2%
Blunder rate
19.4%
P95 equity loss
0.282
Mean loss, worst 5%
0.482
Median / P90 s per decision
0.57 / 0.98
$ per 1,000 decisions
Not measured
Risk judgement, margin 0.001
61.7 (95% interval 58.9 to 64.6); risks found 32.4% (115 / 355), right safety 91.1% (336 / 369)
Risk judgement, margin 0.02
62.6 (95% interval 59.7 to 65.6); risks found 34.3% (109 / 318), right safety 90.9% (369 / 406)
No trade-off
276 (left out of every risk metric)

JSON

Risk call card

61.7risk judgement, margin 0.001

right risks115
wrong risks18
needless risks33
missed risks222
right safety336

Hit risk only: being hit. Gammon risk and priming are not in it.

By game phase

PhaseError rate (PR)Decisions
OpeningToo few decisions143
Contact33.7641
Holding gameToo few decisions20
RaceToo few decisions136
Bear-offToo few decisions60