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

GLM-4.7-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
glm-4.7-flash
Provider or endpoint
our GPUs (vLLM, BF16) 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)
50.645.5 to 56.3
Decisions
1,000 (1,000 scored, 0 invalid)
Engine's best play
36.6%
Blunder rate
33.3%
P95 equity loss
0.464
Mean loss, worst 5%
0.722
Median / P90 s per decision
0.20 / 0.27
$ per 1,000 decisions
Not measured
Risk judgement, margin 0.001
52.4 (95% interval 48.8 to 55.7); risks found 38.9% (138 / 355), right safety 65.9% (243 / 369)
Risk judgement, margin 0.02
52.0 (95% interval 48.5 to 55.5); risks found 38.7% (123 / 318), right safety 65.3% (265 / 406)
No trade-off
276 (left out of every risk metric)

JSON

Risk call card

52.4risk judgement, margin 0.001

right risks138
wrong risks82
needless risks126
missed risks135
right safety243

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

By game phase

PhaseError rate (PR)Decisions
OpeningToo few decisions143
Contact61.3641
Holding gameToo few decisions20
RaceToo few decisions136
Bear-offToo few decisions60