Baidu ERNIE X1 and NVIDIA Llama-Nemotron arrive within two months of R1
Two more labs shipped reasoning models within two months of R1. Baidu unveiled ERNIE X1, positioned as a specialised reasoning model, alongside ERNIE 4.5 on 2025-03-16 (+185 days after o1-preview,…
- Date
- 16 March 2025
- Who
- Baidu, NVIDIA
- Confidence
- Medium (NVIDIA's date is from its model card; Baidu's comes via a Wikipedia poin
- Deep dive
- Reasoning II, from o1 and o3 to DeepSeek-R1 and the labs that replicated them
Tier: Supporting · Significance: 3/5 · Org(s): Baidu, NVIDIA · Confidence: Medium (NVIDIA's date is from its model card; Baidu's comes via a Wikipedia pointer to a PR Newswire release that was not fetchable) Two more labs shipped reasoning models within two months of R1. Baidu unveiled ERNIE X1, positioned as a specialised reasoning model, alongside ERNIE 4.5 on 2025-03-16 (+185 days after o1-preview, +55 after R1; Wikipedia, a pointer). NVIDIA released Llama-3.3-Nemotron-Super-49B-v1 on 2025-03-18 (+187 and +57). It is a 49B model derived from Llama-3.3-70B-Instruct by neural architecture search, post-trained with supervised fine-tuning and then RL (REINFORCE with a leave-one-out baseline, "RLOO," and Online Reward-aware Preference Optimization), with a reasoning on/off switch set in the system prompt, under the NVIDIA Open Model License (model card). NVIDIA's technical report (arXiv 2505.00949, v1 2025-05-02, +232 days) came later than the release; this chapter dates models by their release date throughout. I infer that several labs reached hybrid designs independently, since NVIDIA's prompt-controlled toggle predates Qwen3's /think and /no_think (2025-04-29) by 42 days. Microsoft's Phi-4-reasoning (2025-04-30) and ByteDance's Seed1.5-Thinking (2025-04-10) are in the Diffusion map. Sources: NVIDIA model card · arXiv 2505.00949 · Wikipedia, Ernie Bot