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

GPT-6 Astra

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
gpt-6-astra
Provider or endpoint
ChatGPT subscription (Codex backend)
Date run
Missing
Price
$10.0 / $50.0 per million tokens in / out; API-equivalent (subscription); price date not recorded
Model type
chat model, one letter
Weights
closed
Runs on
subscription
Reasoning setting
reasoning effort 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
not sent (API default)
Notes
answer constrained to the option letters by a JSON schema; stored as choice only

Missing B7 fields: option_groups, price.date, 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)
11.29.5 to 12.9
Decisions
1,000 (1,000 scored, 0 invalid)
Engine's best play
68.7%
Blunder rate
9.8%
P95 equity loss
0.136
Mean loss, worst 5%
0.220
Median / P90 s per decision
4.60 / 10.8
$ per 1,000 decisions
Not measured
Risk judgement, margin 0.001
77.3 (95% interval 74.4 to 80.1); risks found 62.8% (223 / 355), right safety 91.9% (339 / 369)
Risk judgement, margin 0.02
77.8 (95% interval 74.9 to 80.8); risks found 65.7% (209 / 318), right safety 89.9% (365 / 406)
No trade-off
276 (left out of every risk metric)

JSON

Risk call card

77.3risk judgement, margin 0.001

right risks223
wrong risks17
needless risks30
missed risks115
right safety339

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

By game phase

PhaseError rate (PR)Decisions
OpeningToo few decisions143
Contact15.1641
Holding gameToo few decisions20
RaceToo few decisions136
Bear-offToo few decisions60