Temper AI

Temper AI develops compact language models specialized in narrow tasks that run on most consumer PCs and laptops.

Latest

October 10, 2026 Temper Bench v2

150 Rust tasks covering both basic syntax and complex tasks that test understanding of traits, lifetimes, threads and async.

Models

October 7, 2026 Temper 1.5B Beats Qwen2.5-Coder-3B, twice its size, on HumanEval-Rust. Base: Qwen2.5-Coder-1.5B-Instruct October 5, 2026 Temper 0.5B A Rust model that beats models several times its size on Rust. Base: Qwen2.5-Coder-0.5B-Instruct

Other

October 10, 2026 Temper Bench v2 150 Rust tasks covering both basic syntax and complex tasks that test understanding of traits, lifetimes, threads and async. The focus is on practical tasks, and solutions are checked by hidden tests and the compiler. The benchmark is carefully checked for leaks and carries a canary string (canary GUID) that lets it be filtered out of training data.