The 7B Model That Out-Punched Its Weight Class

Released in September 2023 by the newly formed Mistral AI, Mistral-7B-v0.1 matched or exceeded the benchmark performance of Meta's LLaMA 2 13B -- a model nearly twice its size -- while being released under a fully permissive Apache 2.0 license with no usage restrictions at all.

Why it mattered

Mistral-7B became something close to a default starting point for open-source fine-tuning projects over the following year: small enough to run and fine-tune on consumer-grade hardware, strong enough to be genuinely useful, and licensed permissively enough that nobody had to think twice about commercial use. Its architecture choices for efficient long-context inference also influenced a generation of subsequent open model designs.

On the record

• Issuer: Mistral AI
• License: Apache-2.0
• Claimed cutoff: 2023-09-27
• Status: on-chain, attested

See the full hash breakdown and on-chain attestation on Mistral-7B-v0.1's Explorer page.