IplanRIO publishes Rio 3.5 Open 397B on Hugging Face — MIT license, highly competitive scores on terminal + code execution benchmarks, outperforms closed-source competitors including DeepSeek V4 Pro
IplanRIO published Rio 3.5 Open 397B on Hugging Face under MIT license — achieving highly competitive scores on terminal and code execution benchmarks, outperforming closed-source competitors including DeepSeek V4 Pro. The release adds another major open-weight competitor to the H2 2026 open-source coding-and-terminal capability landscape.
The substantive piece is the open-weight beats-closed-source pattern continuation at coding-and-terminal benchmark tier. Pre-Rio-3.5 open-weight coding-and-terminal capability was led by GLM-5.2, MiniMax M3, DeepSeek V4 Pro. Rio 3.5 Open 397B's outperformance of DeepSeek V4 Pro establishes another open-weight option at frontier-tier coding capability — substantively expanding the procurement-options matrix.
The competitive read for the H2 2026 open-weight coding-capability landscape is that multiple vendors continue iterating aggressively. Kimi K2.7 Code HighSpeed 6x throughput + VibeThinker-3B 3B-parameter parity + Rio 3.5 Open 397B together represent the H2 2026 open-weight coding-capability iteration burst across multiple capability + economics dimensions simultaneously.
Hugging Face — Best Open-Source LLM Models in 2026 → · LLM Stats — AI Updates Today (June 2026) →