decision model · v1 set (earlier version)

Perplexity Decider 27B v1.1

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
pplx-decider-v1.1-27b
Provider or endpoint
our GPUs (released checkpoint, its own server code)
Date run
Missing
Price
$0.02 / $0.0 per million tokens in / out; list-price estimate; runs on our GPUs; price date not recorded
Model type
decision model
Weights
open
Runs on
our GPUs
Reasoning setting
none (decision model)
Confidence source
decision-model probability
Rules text
short (the v1 rules text; the same stored request text for every model)
Option groups
groups of up to 25 options (8,192-token limit); no decision in this data was split
Sampling settings
not sent (decision request)
Notes
Missing

Missing B7 fields: 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)
31.827.8 to 36.3
Decisions
1,000 (1,000 scored, 0 invalid)
Engine's best play
49.7%
Blunder rate
22.7%
P95 equity loss
0.302
Mean loss, worst 5%
0.520
Median / P90 s per decision
0.82 / 1.35
$ per 1,000 decisions
Not measured
Risk judgement, margin 0.001
61.6 (95% interval 58.4 to 64.7); risks found 38.9% (138 / 355), right safety 84.3% (311 / 369)
Risk judgement, margin 0.02
61.1 (95% interval 57.7 to 64.3); risks found 38.4% (122 / 318), right safety 83.7% (340 / 406)
No trade-off
276 (left out of every risk metric)

JSON

Risk call card

61.6risk judgement, margin 0.001

right risks138
wrong risks32
needless risks58
missed risks185
right safety311

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

By game phase

PhaseError rate (PR)Decisions
OpeningToo few decisions143
Contact41.2641
Holding gameToo few decisions20
RaceToo few decisions136
Bear-offToo few decisions60