Open Source Model Open Source

gpt-oss-puzzle-88B

A deployment-optimized large language model by NVIDIA, derived from OpenAI's gpt-oss-120b, focused on efficient reasoning and long-context inference.

5/10 MegaOne Score

Visit gpt-oss-puzzle-88B →

Price Free / Open Source

Category Open Source Model

Model Open Source

Company NVIDIA

Team Size 42,000

Total Funding $4.1B

About gpt-oss-puzzle-88B

gpt-oss-puzzle-88B is a deployment-optimized large language model developed by NVIDIA, derived from OpenAI's gpt-oss-120b. It utilizes a post-training neural architecture search (NAS) framework called Puzzle to significantly improve inference efficiency for reasoning-heavy workloads while maintaining or enhancing accuracy. The model is specifically optimized for long-context and short-context serving on NVIDIA H100-class hardware, supporting long-context inference up to 128K tokens.

Why We Scored It 5/10

5/10

The gpt-oss-puzzle-88B model shows technical merit with 2.82x throughput improvements on H100 GPUs, but remains a niche deployment optimization tool rather than a breakthrough model. Despite NVIDIA's backing, it lacks broader market adoption and serves primarily specialized inference efficiency use cases.

Features & Platform

Free Tier

gpt-oss-puzzle-88B Coverage