NVIDIAopen weightsCurrentReasoning

Nemotron 3.5 Bolt 74B-A9B

Name: Nemotron 3.5 Bolt 74B-A9B
Author: NVIDIA

Local-agent model compressed from Super VL and shipped as a 24 GB native 2-bit (IQ2) package for a single consumer GPU; BF16 weights also on Hugging Face. Scores are the 24GB IQ2 build. No NVIDIA list price. NVIDIA reports Terminal-Bench 2.1 51.0 (Terminus 2), PinchBench 87.8.

Released

Oct 15, 2026

Price in / out

—

Blended $/1M

—

Context

256K

Parameters: 74B / 9B active. License: OpenMDW-1.1.

Benchmark scoresvs Nemotron 3 Ultra

Benchmark	Score	Δ vs prior	Rank (current)	Configuration	Source
GPQA Diamond	86.4%	-0.6	#17 of 22	NVIDIA model card, 24GB IQ2 build (86.43) · lab-reported	huggingface.co ↗
Humanity's Last Exam	25.3%	-1.4	#31 of 36	NVIDIA model card, no tools, 24GB IQ2 build; Artificial Analysis run 25 · lab-reported	huggingface.co ↗

Version

$ in / out

Terminal-Bench

SWE-bench

GPQA

HLE

ARC-AGI-2

τ²-bench

Nemotron 3 Ultra

—

50.9%

71.9%

87%

26.7%

—

83.3%

Nemotron 3.5 Bolt 74B-A9B

—

86.4%

25.3%

—

Nemotron 3.5 Super VL

—

71.1%

89.5%

30.7%

1.9%

73.2%

Nemotron 3.5 Bolt 74B-A9B

Benchmark scoresvs Nemotron 3 Ultra

Nemotron versions

Compare

Shipping with it?

Current open-weight peers

Version	Released	$ in / out	Terminal-Bench	SWE-bench	GPQA	HLE	ARC-AGI-2	τ²-bench
Nemotron 3 Ultra	2026-06-01	—	50.9%	71.9%	87%	26.7%	—	83.3%
Nemotron 3.5 Bolt 74B-A9B	2026-10-15	—	—	—	86.4%	25.3%	—	—
Nemotron 3.5 Super VL	2026-10-15	—	—	71.1%	89.5%	30.7%	1.9%	73.2%