Nvidia Launches Nemotron 3.5 Super VL and Two Edge Models

Key Takeaways

1Nvidia launches Nemotron 3.5 Super VL and two open-weight edge models

2Artificial Analysis measures speed and accuracy gains against competitors

3The Western ecosystem accelerates on open models, in a context of security and sovereignty

Why it matters : The arrival of these models could strengthen the position of Western players against Chinese dominance in open-weight models.

With Nemotron 3.5 Super VL and two edge variants, Nvidia is pushing an open-weight offering aimed at agents. Third-party measurements suggest efficiency gains, while the Western ecosystem is buzzing in response to historical Chinese models.

Western players ramp up releases against Chinese models

The geographical gap could narrow in open-weight models. The French lab Mistral has launched Le Chonk, which has reached certain peak benchmarks. Nvidia has invested in Reflection AI and is in talks to acquire it, while Reflection AI released its Beam model last week, generating excitement. Historically, Chinese companies like DeepSeek and Moonshot.ai have offered cheaper and more customizable open models compared to closed ones. This dynamic is part of a broader shift in the AI landscape, with a rising Western presence challenging the past dominance of Chinese labs.

Security and sovereignty drive open-weight adoption

The adoption of open-weight models is intensifying following a security breach of Hugging Face by OpenAI, which dispelled the notion that only open models carry risks. If open models produced in the United States demonstrate their value against Chinese offerings, they could make headway in sectors still hesitant, given cybersecurity recommendations that are unfavorable to foreign technologies.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Two edge variants accompany Super VL for local uses

Two edge models, Nemotron 3.5 Bolt 74B-A9B 24GB and Nemotron 3.5 Bolt E4B 4GB, complement the announcement. Nvidia presents the first as intended for local agentic uses and the second for fast conversational agents. Edge AI encompasses models capable of operating locally and offline. Meanwhile, Nvidia is making Nemotron 3.5 Super VL available starting today.

Claimed performance: speed, accuracy, and limitations of the base model

Nvidia presents Nemotron 3.5 Super VL as a model for specialized agents, combining high processing speed and accuracy. The company claims it is five times faster than the next best in its category, based on tests from Artificial Analysis, where InclusionAI's Ling 3.0 Flash and DeepSeek's V4.1 Flash ranked second and third, respectively. Nvidia highlights top-tier accuracy thanks to post-training, citing a 44% gain in a medical insurance use case, where the model performed well above Claude Opus 5. However, the base model alone has not surpassed leading proprietary models. Measurements from Artificial Analysis indicate efficiency gains for now, and the model includes 120 billion parameters.

Openness and data control, key arguments put forth by Nvidia

Nvidia argues that open models can be fine-tuned on internal data and deployed on controlled infrastructure for greater customization and security. However, it remains uncertain whether the new model truly raises the bar for open or proprietary models in practice.