The compact model, post-trained from Google's Gemma 4 E4B, targets GeForce RTX GPUs and Jetson edge devices
NVIDIA Corporation (NVDA) price
Full chart and statsLoading chart…
Market cap $5.3T (shares as of 2026-08-21)
P/E (trailing) 27.6 (EPS to 2026-07-26)
Annual operating margin 60.4% (FY ended 2026-01-25)
Forward P/E Unavailable: no dated forecast source
Gloria daily close for session Oct 27, 2026. Market cap and trailing P/E use this close with the SEC share total and trailing diluted earnings, each dated next to its figure. Annual revenue, revenue growth, net income and margin show their own fiscal period and link the 10-K. Current context may postdate this article.
NVIDIA Corporation and its peers
Token vs underlying stock
Current comparison for NVDA. Historical observations may postdate this article.
NVDAx token unit vs NVDA share · USD
$218.06Oct 27 UTC
+2.7% over displayed period
● NVDAx token┄ NVDA stock: $217.94
Nvidia has released the Nemotron 3.5 Bolt E4B, a 4GB model built to run voice assistants and AI agents directly on local hardware.
A small model with big ambitions
The Bolt E4B launched on October 23, 2026. It is the newest member of the Nemotron 3.5 family, a lineup aimed at AI agents that run on-device rather than in the cloud.
Nvidia designed it for low-memory hardware. The target list includes GeForce RTX GPUs and Jetson edge devices, with the model tuned for workloads that fit inside 4GB of VRAM.
The model was not built from scratch. It is a post-trained configuration of Google’s Gemma 4 E4B, meaning Nvidia took Google’s open base model and refined it for its own purposes.
On paper, the Bolt E4B carries 8 billion total parameters and 4 billion effective parameters. It also supports a 64K context window.
The numbers Nvidia is pointing to
Nvidia’s headline claim is performance. The company says the Bolt E4B yields approximately 40% higher average benchmark scores than its predecessor, the Nemotron 3 Nano 4B.
The other claim is about efficiency. Nvidia says the model reduces output tokens by up to 80% in specific configurations.
The model is fine-tuned for efficient conversation and multi-turn tool use. It also supports asynchronous tool calling, so an agent can kick off a task and keep talking rather than freezing while it waits for a result.
The Bolt E4B supports 35 languages. Nvidia is pitching the model at voice assistants, interactive characters, and robotics interfaces.
Part of a tiered lineup
The Bolt E4B did not arrive alone. On October 15, Nvidia announced two larger siblings in the same family.
The first is the Nemotron 3.5 Super VL, with 120 billion total parameters. The second is the Nemotron 3.5 Bolt 74B-A9B, which runs with 24GB of memory.
Nvidia is also keeping the release open. The weights, datasets, and training recipes are available under the OpenMDW-1.1 license.
Developers can access the model through Hugging Face, llama.cpp, and vLLM.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

3 weeks ago
29








English (US) ·