Meta (Muse Spark Open), NVIDIA (Nemotron 3.5) and Reflection AI (Beam 1.0) all released open-weight models in mid-October. The highest-scoring of them, Muse Spark Open, scores 25 on the Artificial Analysis Index v5; the highest-scoring open models overall, both from Chinese labs, score 32.
More choice for businesses that need to consider model origin, especially at the small, fast end, which is also the cheapest to run on your own hardware.
What happened
- 15 Oct: Meta released Muse Spark Open (816B total, Mixture of experts (active parameters): A design where only part of the model (the active parameters) works on each piece of text. Glossary, multimodal), the highest-scoring open model from a US lab on AA v5 to date.
- 15 Oct: NVIDIA released Nemotron 3.5 Super VL (120B / 12B, vision, 1M context) and Nemotron 3.5 Bolt 74B-A9B, a light agent model.
- 16 Oct: Reflection AI released Beam 1.0 (501B / 23B, text) under Apache-2.0.
How they score
On the Artificial Analysis Intelligence Index v5: Muse Spark Open 25, Reflection Beam 1.0 18, Nemotron 3.5 Super VL 14. The highest-scoring open models, DeepSeek V4.1-Flash-1017 and GLM-5.3 (both China), score 32. On LMArena's open-source text ranking, the top five open models come from Chinese labs (1476-1488); Google's Gemma 4 31B (US) scores 1452.
OpenAI and Google have released no new open LLM (large language model): An AI model trained on large amounts of text to read, summarise and write language. Glossary since gpt-oss (August 2025) and Gemma 4 (April 2026).
Licences
- Muse Spark Open: Meta's own licence, with a "Built with Muse Spark Open" attribution notice; model-as-a-service offerings above $20M need an agreement.
- Reflection Beam 1.0: Apache-2.0.
- Nemotron 3.5: OpenMDW-1.1 (permissive).
Our view
Several of the new US-origin releases are small and efficient (Gemma 4, Recursive Flint, Nemotron Bolt), which is also the cheapest class to run locally. For organisations whose contracts or policies restrict model origin, that widens the practical choice. As always, we list each model's origin so you can apply your own policy. Full comparison in the Open-weight model tracker.
Why it matters: More choice at the small, efficient end, which is the cheapest class to run in-house.
Source: Meta AI Research, “Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date”, 15 Oct 2026