Code Llama
Meta's code-specialized Llama 2 family — the template every later open coder competed against.
About
Code Llama (August 2023) is Meta's code-focused fine-tune of Llama 2, trained on 500B more tokens of code, in 7B/13B/34B sizes (a 70B followed in January 2024), each with base, Python-specialized, and Instruct flavors. Its context stretched from 4K to 100K tokens, which made whole-file editing practical. Released under the Llama 2 Community License — commercial use allowed for most companies, but it's a custom Meta license, not open source. Benchmarks have long been surpassed, but its ecosystem (WizardCoder, Phind-CodeLlama fine-tunes) kept it relevant for years.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Meta-reported: Code Llama 34B — HumanEval 53.7 pass@1, roughly GPT-3.5 level at release; 70B — HumanEval 53, MBPP 62.4.
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: Meta AI blog + research publication read this session: Aug 24 2023 release, 7B/13B/34B + 70B, Llama 2 Community License, 100K context, HumanEval 53.7 (34B).; Deep verification: multiple corroborating sources (best-of-ai catalog, CoinTelegraph, Gadgets360). License caveat (custom, non-OSI) explicit..