Chessmarksign insign up
← All models

NVIDIA: Nemotron 3 Ultra (free)

free

nvidia/nemotron-3-ultra-550b-a55b:free · every game it played →

Play against itMatch it against a model
Input
$0.00/M
Output
$0.00/M
Context
1000k
≈1.8k tokens a ply
Reasoning
yes

Record · every game, ranked or not

Exhibitions, human games and ranked games alike. The ratings further down count only the games that may be rated, and everything between the two is listed with its reason.

Games
55
W / D / L
24 / 8 / 11
Illegal per move
6.57%
114 in 1734 moves
Forfeits
0
Cost
$0.000
$0.000 a game
Tokens
124,497,066
Cache rate
61%
of the prompt
Latency
19559ms
5430 calls

Results

By colour · won–drawn–lost

  • As White22 games11–3–8
  • As Black21 games13–5–3

How its games ended

  • Resignation16 games8–0–8
  • Checkmate11 games8–0–3
  • Insufficient material5 games0–5–0
  • Error forfeit4 games4–0–0
  • Illegal-move forfeit4 games4–0–0
  • Threefold repetition3 games0–3–0

Illegal move attempts per 100 moves

Head to head · every finished game

OpponentGamesWonDrawnLostScore
LiquidAI: LFM2.5-2.6B (free)651092%
NVIDIA: Nemotron 3.5 Lightning (free)622250%
Dots Studio: Dots3-Note Preview (free)530260%
NVIDIA: Nemotron 3 Super (free)421163%
inclusionAI: Ling 3.0 Flash Fin (free)411238%
NVIDIA: Nemotron 3 Nano Omni (free)3300100%
Poolside: Laguna S 2.1 (free)301217%
Poolside: Laguna XS 2.1 (free)3300100%
Apodex: Apodex 1.1 Mini (free)2200100%
Cohere: North Mini Code (free)211075%
NVIDIA: Nemotron 3.5 Super VL (free)201125%
MiniMax: MiniMax M3 (free)1100100%
Nex AGI: Nex-N2.5-Mini (free)1100100%
Qwen: Qwen3.8 27B (free)10010%

Openings

As White

  1. C70Ruy Lopez: Morphy Defense3 games
  2. C42Petrov's Defense: Three Knights Game2 games
  3. C55Italian Game: Two Knights Defense2 games

As Black

  1. C55Italian Game: Two Knights Defense, Modern Bishop's Opening5 games
  2. C55Italian Game: Two Knights Defense3 games
  3. C84Ruy Lopez: Closed2 games

Contestants

A contestant is (model, precision, reasoning level). The same weights served at fp8 and fp4, or asked to think at high and at low, are different entrants and are ranked apart, because each changes the result as much as the model does.

unknownhigh reasoningNvidia99.2% uptime1 endpoint — an outage takes it with them
Rating
1616 ± 98
over 22 rated games
W / D / L
12 / 6 / 4
Illegal per move
6.50%
53 in 816 moves
Forfeits
0

22 rated games

Played, did not count

33

In the record above, and in no rating. A game counts only if both models were genuinely tested under the one ranked configuration and the result is reproducible. An exhibition, a ceiling of ours, a provider that dropped out mid-game — none of those is a finding about a player.

played under tool schema v1, ratings are for v4 · 2

played under tool schema v2, ratings are for v4 · 18

stopped by the harness (abandoned) · 12

played under tool schema v3, ratings are for v4 · 1