chat model, one letter · League 2
Gemini 3.8 Flashhigh
FinishedPublished
Results f24d1f41, updated 16 November 2026, 21:24 UTC; scored by GNU Backgammon 1.08, 2-ply.
How we called it
- Model ID
- gemini-3.8-flash-high
- Provider or endpoint
- Google Antigravity subscription
- Date run
- 2026-10-19 to 2026-10-26
- Price
- $0.75 / $3.75 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
- thinking level high
- Confidence source
- none
- 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 237 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
- Google Antigravity subscription from 19 Oct 12:46 UTC (Antigravity model id gemini-3.8-flash-high, model version gemini-3.8-flash); before that it waited for a Gemini API key; answer letter constrained to the option letters
Missing B7 fields: price.date
Results in League 2
- Error rate (PR)
- 14.912.9 to 17.0
- Decisions
- 2,168 (2,168 scored, 0 invalid)
- Engine's best play
- 63.1%
- Blunder rate
- 11.0%
- P95 equity loss
- 0.167
- Mean loss, worst 5%
- 0.305
- Median / P90 s per decision
- 9.59 / 44.4
- $ per 1,000 decisions
- 22.352API-equivalent (subscription)
- Risk judgement, margin 0.001
- 66.9 (95% interval 65.2 to 68.5); risks found 36.1% (264 / 731), right safety 97.6% (858 / 879)
- Risk judgement, margin 0.02
- 67.6 (95% interval 65.5 to 69.6); risks found 39.2% (239 / 610), right safety 96.1% (961 / 1,000)
- No trade-off
- 558 (left out of every risk metric)
Risk call card
66.9risk judgement, margin 0.001
right risks264
wrong risks17
needless risks21
missed risks450
right safety858
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
- 2,168
- Eligible (full probability distribution)
- 0
- Left out, by reason
- not audited 194; one letter only, or a program 1,974
- Close choices
- N/A no full probability distribution
By game phase
| Phase | Error rate (PR) | Decisions |
|---|---|---|
| Opening | 12.0 | 308 |
| Contact | 20.2 | 1,363 |
| Holding game | Too few decisions | 54 |
| Race | 2.0 | 244 |
| Bear-off | Too few decisions | 199 |