OpenThinker-32B
Open-data reasoning model built on 114K verified traces
About
OpenThinker-32B is a Qwen2.5-32B-Instruct fine-tune from the OpenThoughts consortium (researchers at Stanford, Berkeley, UCLA, Juelich, and Toyota Research Institute). Its differentiator is data transparency: unlike DeepSeek's private training data, the full OpenThoughts-114k dataset of verified reasoning traces is public, built by collecting attempts from DeepSeek-R1 over 173K questions and filtering for correctness. The team reports it beats DeepSeek-R1-Distill-Qwen-32B on MATH500 and GPQA-Diamond. The 16K context window is modest for a reasoning model, and coding scores trail the DeepSeek distill slightly.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Team-reported: MATH500 90.6, GPQA-Diamond 63.5, AIME 2024 68.0, LiveCodeBench 68.6
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: OpenThoughts blog read (Feb 12 2025 release): Qwen2.5-32B-Instruct base, OpenThoughts-114k verified dataset, 16K context, Apache 2.0.; Benchmarks from team: MATH500 90.6, GPQA-Diamond 63.5, AIME24 68.0 — team-reported via Evalchemy.; Consortium backing: Stanford, Berkeley, UCLA, Juelich, Toyota Research Institute.; Limitation: coding scores slightly trail R1 distill; 16K context modest for reasoning..