chat model, answer-letter probabilities · League 2

Qwen3.8-27Blow · NVIDIA RTX PRO 6000 Blackwell + NVIDIA DGX Spark

FinishedPublished

Results f24d1f41, updated 16 November 2026, 21:24 UTC; scored by GNU Backgammon 1.08, 2-ply.

How we called it

Model ID
qwen3.8-27b
Provider or endpoint
NVIDIA RTX PRO 6000 Blackwell (96 GB), fp16 (vLLM); first decisions on Cerebras via OpenRouter; NVIDIA DGX Spark (GB10) from 20 Oct 21:10 UTC (back on the NVIDIA RTX PRO 6000 Blackwell (96 GB) 21 Oct 01:25-05:00)
Date run
2026-10-19 to 2026-10-26
Price
no price per token; estimate: GPU time, at most; source content/own_cost.json (wall clock on our machines × hourly rate); price date not recorded
Model type
chat model, answer-letter probabilities
Weights
open
Runs on
our GPUs
Reasoning setting
thinking on, reasoning effort low (the chat template's reasoning_effort)
Confidence source
letter probability
Rules text
long
Option groups
up to 26 options per request (letters A-Z); more legal plays are split into equal groups, a winner per group, then among the winners; in this data 217 decisions were split into groups of 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26
Sampling settings
temperature 1.0
Notes
Cerebras via OpenRouter (model ID qwen/qwen3.8-27b, fp16) before 19 Oct 22:26 UTC, where it finished no match (its first 4 decisions came from there); then on our GPU (vLLM, fp16); letter read from a prefilled Answer: after the reasoning
Earlier name
Qwen3.8-27B (on) (renamed on 28 Oct 2026 to the reasoning effort its requests sent)

Results in League 2

Error rate (PR)
29.026.5 to 31.8
Decisions
2,108 (2,108 scored, 0 invalid)
Engine's best play
52.8%
Blunder rate
22.4%
P95 equity loss
0.290
Mean loss, worst 5%
0.463
Median / P90 s per decision
171.1 / 566.7
$ per 1,000 decisions
data-v="39.023">39.023estimate · RTX PRO 6000 Blackwell + API
Risk judgement, margin 0.001
62.0 (95% interval 60.1 to 63.9); risks found 33.8% (242 / 715), right safety 90.2% (735 / 815)
Risk judgement, margin 0.02
61.2 (95% interval 59.2 to 63.2); risks found 34.4% (217 / 630), right safety 87.9% (791 / 900)
No trade-off
578 (left out of every risk metric)

JSON

Risk call card

62.0risk judgement, margin 0.001

right risks242
wrong risks56
needless risks80
missed risks417
right safety735

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

Close choices

A decision is a close choice when the second most likely play has at least half the probability of the first (internally: split).

Probabilities over plays, not sampled reasoning. Method

League 2: complete letter-readout records are mostly few-option positions, a selected subset.

Decisions
2,108
Eligible (full probability distribution)
298
Left out, by reason
asked in groups of options 142; incomplete distribution (an option without its own probability) 939; not audited 729
Close choices
0.0% (0 / 298)
Ties in the top two
0 decisions (0 at a positive probability, 0 at 0; the two most likely plays tied: 0, the second tied with the third: 0); 0 of them close choices. Ties are broken by the order the plays were shown, never by the engine.
Between a safe play and a riskier one
N/A (0 / 0) of the close choices; 0.0% (0 / 298) of all eligible decisions
Probabilities of the two most likely plays, mean with median
DecisionsNumberp1 (most likely play)p2 (second)p1 + p2
Close choices0N/AN/AN/A
Other decisions2981.000median 1.0000.000median 0.0001.000median 1.000
How much probability the two plays hold together: decisions per bin
p1 + p20.0–0.10.1–0.20.2–0.30.3–0.40.4–0.50.5–0.60.6–0.70.7–0.80.8–0.90.9–1.0
Close choices0000000000
Other decisions000000000298
What differs between the two most likely plays (several can differ; none is picked as the cause)
DecisionsNumberHitting rollsNewly made pointsPip countsLongest primeNone / 1 / 2 / 3 / 4 differ
Close choices0N/A0 / 0N/A0 / 0N/A0 / 0N/A0 / 00 / 0 / 0 / 0 / 0
Other decisions29835.2%105 / 29815.4%46 / 29813.1%39 / 29830.2%90 / 29896 / 138 / 50 / 14 / 0
Engine values (GNUbg 2-ply, equity), mean with median
DecisionsNumberGap between the two plays' valuesMost likely play: behind the engine's bestSecond most likely play: behind the engine's best
Close choices0N/AN/AN/A
Other decisions2980.120median 0.0540.031median 0.0000.112median 0.040
Close choices between a safe play and a riskier one, by position class
MarginIn Risk pays positionsIn Safety pays positions
margin 0.001N/A0 / 0N/A0 / 0
margin 0.02N/A0 / 0N/A0 / 0
Close choices among eligible decisions, by position class and stake band
MarginRisk paysSafety paysNo trade-offRisk pays, small stakeRisk pays, medium stakeRisk pays, big stake
margin 0.0010.0%0 / 33, under 200 decisions0.0%0 / 90, under 200 decisions0.0%0 / 175, under 200 decisions0.0%0 / 3, under 200 decisions0.0%0 / 14, under 200 decisions0.0%0 / 16, under 200 decisions
margin 0.020.0%0 / 30, under 200 decisions0.0%0 / 93, under 200 decisions0.0%0 / 175, under 200 decisionsN/A0 / 00.0%0 / 14, under 200 decisions0.0%0 / 16, under 200 decisions

At the 0.02 margin a Risk pays position has a stake of at least 0.02, so the small band is empty. A big stake means playing safe costs more by the engine's value, not that the decision was harder.

Missed risks where the risk is worth taking, close choices vs other decisions
MarginClose choices: missed risksOther decisions: missed risks (weighted to the close choices' mix)Difference, close minus otherCrude difference
margin 0.00124 decisionsN/AN/A (0 / 0)70.8%70.8% (17 / 24), under 200 decisionsN/A: no close choices in this groupN/A
margin 0.0223 decisionsN/AN/A (0 / 0)69.6%69.6% (16 / 23), under 200 decisionsN/A: no close choices in this groupN/A

By game phase

PhaseError rate (PR)Decisions
Opening32.9288
Contact34.21,305
Holding gameToo few decisions82
Race15.5246
Bear-offToo few decisions187