- Score · relative 0–100
- 21.9
- Rank
- 67
- Status
- Partial 3/3
- Players generated
- 2026-07-02 – 2026-07-11
- Match evidence updated
- 2026-10-06
- Results published
Why this result is partial: one or more documented settings are untested or ineligible. The rank uses the available reasoning-setting results.
Scores by reasoning setting
Available settings and grouping
Available settings
Checked on 2026-07-30 · Provider documentation
How settings are grouped
- Baseline: Reasoning disabled — available, not tested in the current results
- Balanced: Medium — tested and included
- Intensive: High — tested and included
Scores by game
What this result shows
Strongest included setting: Medium (23.6).
Across included settings, the highest game score is International Draughts (40.4); the lowest is Tumbleweed (18.2). These are relative scores within each game, not direct comparisons of difficulty.
Match evidence measures these retained programs. It does not establish how a fresh generation will perform. Independent-generation variation is not estimated in this release.
Test protocol and evidence limits
- Model identifier
- nvidia/nemotron-3-ultra-550b-a55b
- Provider route (catalog)
- openrouter
- Prompt identity / profile
- Not retained in this result set / Not retained in this result set
- Prompt version
- Not retained in this result set
- Generation policy
- Generation-quality admission policy (see methodology)
- Requested output-token limits
- Not retained (0 of 56 indexed attempts)
Generation evidence
- Medium: 5 contributing player programs across 5 games
- High: 9 contributing player programs across 6 games
Counts describe contributing programs, not independent replications: repairs and resends are not new independent samples. Independent-generation lineage is not retained; the per-game counts below keep unique code identities separate from admitted generation runs.
| Game | Setting | Unique programs | Admitted generation runs | Program score range |
|---|---|---|---|---|
| Chess | high | 2 | 2 | 6.4–8.3 |
| Chess | medium | 1 | 1 | 7.0–7.0 |
| Game of the Amazons | medium | 1 | 1 | 12.6–12.6 |
| Game of the Amazons | none | 1 | 1 | 4.4–4.4 |
| Go | high | 1 | 1 | 24.1–24.1 |
| Go | medium | 1 | 1 | 42.9–42.9 |
| Go | none | 1 | 1 | 45.2–45.2 |
| Hex | high | 2 | 2 | 27.7–37.0 |
| Hex | none | 1 | 1 | 52.2–52.2 |
| International Draughts | high | 1 | 1 | 38.9–38.9 |
| International Draughts | medium | 1 | 1 | 56.0–56.0 |
| International Draughts | none | 1 | 1 | 26.2–26.2 |
| Shogi | none | 1 | 1 | 12.8–12.8 |
| Tumbleweed | high | 2 | 2 | 17.6–18.8 |
| Tumbleweed | medium | 1 | 1 | 18.3–18.3 |
| Tumbleweed | none | 1 | 1 | 18.2–18.2 |
| Xiangqi | high | 1 | 1 | 10.7–10.7 |
| Xiangqi | none | 1 | 1 | 12.8–12.8 |
Program ranges describe the retained competitors on this release scale, not a statistical estimate of a fresh generation.
Failures and budgets
13 model-code generation failures; 0 provider/infrastructure failures excluded from the generation denominator. Runtime player faults are separate: 0 recorded unilateral faults across contributing program rows, excluded from rating evidence.
Repair/resend outcomes and available costs are included. Exact retry ceilings and execution budgets are not retained in this result set; see the methodology for the public policy. Do not infer a measured setting from a current provider catalog.
Player program results
High: 12% / 44% / 44% · n 16Medium: 0% / 56% / 44% · n 9Reasoning disabled: 56% / 22% / 22% · n 9 All three benchmark settings testedMany programs needed repairMany programs failedGeneration cost
$0.03average estimated cost per game and reasoning setting
- Median
- $0.03
- Range
- $0.01–$0.13
- Known cost subtotal (incomplete or unverified charges)
- $1.21
- Cost data
- 24 combinations · 34 runs
- Price source
- Provider-reported
- Price list
- Not recorded for this older data
- Average output
- 7.8k tokens
- Output range
- 2.7k–18k tokens
- Runs with output-token data
- 25
How cost is measured
Comparison cost gives equal weight to each model, game, and reasoning-setting combination. Recorded generation attempts include repairs and resends when cost evidence is available; runtime compute and service operation are excluded. Cost and output-token counts use separate telemetry denominators. A list-price estimate is not actual expenditure; mixed evidence is not a uniform standard-price comparison.
Test protocol
- Combines this model's tested reasoning settings
- Player-program generation; retained protocol details are under “Test protocol and evidence limits” above