nvidia/nemotron-3-ultra-550b-a55b
nvidia/nemotron-3-ultra-550b-a55bSamples: 16
Open, efficient hybrid Mamba-Transformer MoE with 1M context for agentic workflows.
- Per request
- $0.00
- Context window
- 262,144
- Route
- Default
DefaultRoleplay combines public evaluations of emotional understanding, creative writing and story-related skills. It is a comparative reference, not a direct roleplay test.
Each benchmark is rescaled to 1–10 within its own leaderboard. We average results from the same publisher first, then give each publisher equal weight. Only verified model matches are included.
LiveBench · Story constraint following
nemotron-3-ultra-550b-a55b: 69.23
71 Models · 2026-10-25
LiveBench · Theory of mind
nemotron-3-ultra-550b-a55b: 61.54
71 Models · 2026-10-25
Routes
Ability score divided by token cost, ranked across measured model routes. Cost uses equal input and output token counts; higher is better value.
| Use this model | ||||||
|---|---|---|---|---|---|---|
| Default | $0.00Per request | — | — | 4.9/10 | Not measured | Use this model |
Quick start
Access and billingcurl --fail-with-body 'https://api.tokenreply.com/v1/chat/completions' \
-H "Authorization: Bearer $TOKENREPLY_API_KEY" \
-H 'Content-Type: application/json' \
--data '{
"model": "nvidia/nemotron-3-ultra-550b-a55b",
"messages": [
{
"role": "user",
"content": "Reply with hello."
}
]
}'