Nemotron 3 Ultra
Model ● Covered in 7 stories + Follow
Nemotron 3 Ultra is a 550-billion-parameter open model from NVIDIA with a 1-million-token context window that has demonstrated strong performance on code and agentic AI tasks. The model achieved 71.7% on SWE-bench for fixing real software bugs and top accuracy on LangChain's Deep Agents benchmark, while running at significantly lower inference costs than closed models. It has become a benchmark reference point for evaluating other models' performance on coding and agent-based workloads.
Updated 4 August 2026
Specifications
No specifications recorded yet.
Latest developments
Q3 2026
Salesforce debuted Koa, a CRM-focused reasoning model built on Nvidia’s Nemotron, for use in its Agentforce platform Model release
Thinking Machines Lab releases Inkling, a 975B-parameter open-weights multimodal mixture-of-experts model Open source release
Thinking Machines Lab releases Inkling, a 975-billion-parameter open-weight AI model Open source release
- Salesforce Builds Its First AI Model on Nvidia’s Open-Weight Model Nemotron
- NVIDIA Vera Rubin Maximizes Intelligence per Dollar for Post-Training Workloads – a Key Metric for Agentic AI
- Thinking Machines Lab Releases Inkling: A 975B-Parameter Open-Weights Multimodal MoE With 41B Active Parameters And Controllable Thinking Effort
- Thinking Machines amps up its bet against one-size-fits-all AI with its first open model, Inkling
- NVIDIA Nemotron Achieves Benchmark-Leading Performance With LangChain Deep Agents Harness
Q2 2026
Anthropic releases Claude Fable 5 model with improved benchmarks and new safety restrictions Model release