Subscribe
Developers6 min read

What are open-weight models, and how far behind the frontier are they?

Downloadable parameters under a licence, not open source, currently landing at 58 percent of the leading closed model's score for 44 percent of the cost per point.

Read the brief · 45 seconds
· The Open Source AI Definition requires data information, complete training code and parameters, and says open source weights must include the data information and code used to derive them· DeepSeek V4.1-Flash is a 552B-backbone mixture of experts activating 8B parameters per token during prefill and 16B during decode, trained on a 45T-token multimodal corpus, released under MIT· OpenBMB released MiniCPM5-2B under Apache 2.0 with its training datasets and GGUF builds for llama.cpp, Ollama and LM Studio· MiniCPM5-2B is a dense 2.5-billion-parameter model built for on-device deploymentGo deeper · 6 min
Published by TheSampleSpace · Sep 11, 2026

Developers

VIEW ALL »

Hardware

VIEW ALL »

Business

VIEW ALL »
The Sample · Newsletter

You just read today in 30 seconds. Get it in your inbox.

One lead story, three briefs, one number. Monday, Wednesday and Friday.

Free. Every claim sourced. Unsubscribe in one click.