A burst of open-weight model releases in the first week of May 2026: Mistral Medium 3.5 (128B unifying chat, reasoning, and coding), Poolside's Laguna XS.2 (68.2% SWE-bench on a single GPU), Gemma 4 Multi-Token Prediction drafters for faster inference, and xAI's Grok 4.3 topping agentic tool-calling benchmarks.
xAI has released Grok 4.3 on its API, claiming top spots on agentic tool calling and instruction-following leaderboards, and ranking #1 in enterprise domains such as case law and corporate finance. It supports a 1M token context window at $1.25/M input and $2.50/M output.
Google has released Multi-Token Prediction (MTP) draft models for the Gemma 4 family, achieving up to 3x inference speedup through speculative decoding without any loss in output quality.
Poolside AI released Laguna XS.2 on April 28, 2026 under Apache 2.0 — a 33B total/3B active MoE model purpose-built for agentic coding, scoring 68.2% on SWE-bench Verified and deployable on a single consumer GPU.
Released April 29, 2026 under Modified MIT license, Mistral Medium 3.5 consolidates the company's chat, reasoning, and coding models into one 128B dense open-weight model with 256K context, scoring 77.6% on SWE-bench Verified.