Nemotron 3.5 Super VL - Thorsten Meyer AI

Reality Check

Nemotron 3.5 Doubles NVIDIA’s Score. America’s Open-Weight Gap Is Still There.

A large and a compact glowing cobalt engine on a bright white floor, with turquoise bars rising toward the horizon

NVIDIA’s Nemotron 3.5 Super VL scores 26, twice its predecessor, at $0.25 per index task. Bolt nearly matches June’s 550B flagship with 9B active parameters. The best permissive US open model still ranks 18th.

Thorsten Meyer
October 16, 2026

View Post

Reality Check

Muse Spark Open: America’s Strongest Open-Weight Model, Priced Like Meta’s Closed One

Open glass display case holding a glowing cluster of cobalt and turquoise crystals threaded with luminous wires, against a bright sky

Muse Spark Open is the best US open-weight model by 12 points and leads all open models on knowledge reliability. It also costs the same per token as the closed Muse Spark 1.3, scores 8% on Terminal-Bench 4.0 and ships under a license you need to read.

Thorsten Meyer
October 16, 2026

View Post

Reality Check

Nemotron 3.5 Bolt: An Ultra-Class Score at a Seventh of the Size. You Pay for It in Tokens.

Small glowing glass cube beside a tall dark-blue server rack, its light beam rising into a bright blue sky

NVIDIA’s Nemotron 3.5 Bolt 74B-A9B scores 21.4, close to the 550B Nemotron 3 Ultra. It is the best US model under 100B parameters, but fourth in its class, and it writes 82,000 tokens per task.

Thorsten Meyer
October 16, 2026

View Post

Reality Check

Nemotron 3.5 Super VL: 25 Cents a Task, Sub-Second Latency, and a Score of 26

Translucent blue turbine engine in a bright sky, emitting fast streams of turquoise light

NVIDIA’s Nemotron 3.5 Super VL doubles its predecessor’s score, costs $0.25 per index task and answers fast. But at 26 points, open models with MIT licenses deliver far more intelligence for the same money.

Thorsten Meyer
October 16, 2026

View Post