Nemotron 3 Super 120B A12B
thinkingmachines/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 · family nemotron · in: text → out: text · price provider-declared (via models.dev). Capabilities: tools reasoning text.
$0.79
Blended $/1M (3:1)
$0.57
Input $/1M
$1.44
Output $/1M
$2.01
1M in + 1M out
$0.114 −80%
Cache read $/1M
#3945 of 8,512
Cheapest rank
- Provider
- Thinking Machines (
thinkingmachines) - Kind
- text (outputs text)
- Context
- 65,536 tokens
- Max output
- 65,536 tokens
- Capabilities
- tools reasoning text
Same model, other providers
“Nemotron 3 Super 120B A12B” is offered by 8 providers — cheapest-effective first. Requesty is the cheapest at $0.00 blended.
| Provider | Input | Output | Blended | Context |
|---|---|---|---|---|
| Requesty | free | free | $0.00 | 1M |
| OpenRouter | $0.075 | $0.4 | $0.16 | 1M |
| Kilo Gateway | $0.075 | $0.4 | $0.16 | 262k |
| Kenari | $0.0971 | $0.48 | $0.19 | 262k |
| Boundless | $0.12 | $0.52 | $0.22 | 262k |
| Synthetic | $0.3 | $1 | $0.47 | 262k |
| Thinking Machines (this) | $0.57 | $1.44 | $0.79 | 66k |
| Crusoe | $0.3 | $2.4 | $0.82 | 262k |
Call it
jishie indexes & prices models; it does not proxy inference.
Most providers are OpenAI-compatible — point base_url at Thinking Machines and pass this model id:
curl $BASE_URL/v1/chat/completions \
-H "Authorization: Bearer $API_KEY" -H "Content-Type: application/json" \
-d '{"model":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","messages":[{"role":"user","content":"hello"}]}'
Route programmatically with MCP find_model / get_model, or fetch this record free (no wallet):