chat model, one letter · v1 set (earlier version)

Gemini 3.8 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
gemini-3.8-flash
Provider or endpoint
Gemini API
Date run
Missing
Price
$0.75 / $3.75 per million tokens in / out; API price; price date 2026-10-14
Model type
chat model, one letter
Weights
closed
Runs on
API
Reasoning setting
thinking level low
Confidence source
none
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
no token probabilities from this API; answer constrained to the option letters, stored as choice only

Missing B7 fields: 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)
18.316.0 to 20.7
Decisions
1,000 (1,000 scored, 0 invalid)
Engine's best play
58.6%
Blunder rate
15.8%
P95 equity loss
0.193
Mean loss, worst 5%
0.298
Median / P90 s per decision
2.87 / 7.96
$ per 1,000 decisions
Not measured
Risk judgement, margin 0.001
66.4 (95% interval 63.8 to 69.1); risks found 36.6% (130 / 355), right safety 96.2% (355 / 369)
Risk judgement, margin 0.02
67.5 (95% interval 64.7 to 70.3); risks found 39.6% (126 / 318), right safety 95.3% (387 / 406)
No trade-off
276 (left out of every risk metric)

JSON

Risk call card

66.4risk judgement, margin 0.001

right risks130
wrong risks13
needless risks14
missed risks212
right safety355

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

By game phase

PhaseError rate (PR)Decisions
OpeningToo few decisions143
Contact23.6641
Holding gameToo few decisions20
RaceToo few decisions136
Bear-offToo few decisions60