Meta ships Llama 5.1 405B — open-weights flagship returns to monthly cadence, reasoning-tier eval gaps with closed-flagships narrow further
Meta released Llama 5.1 405B this week, a six-week iteration on Llama 5.0 405B from early April. The 5.1 weights ship under the modified Llama Community License and close several measured gaps with closed-flagship models on reasoning evals — MATH from 0.78 to 0.84, GPQA-Diamond from 0.61 to 0.69, SWE-bench Verified from 0.49 to 0.57.
The eval gap with the closed flagships is now narrower than it has been since the original Llama 3 405B. On SWE-bench Verified specifically, Llama 5.1 405B's 0.57 score puts it within 12 points of Claude 4.7 Opus's 0.69 — a margin that closes once Llama 5.1 fine-tunes specialized for agent-IDE use enter the ecosystem.
The faster cadence is the strategic move. Meta is committing to a monthly Llama refresh through 2026, with major-version weight drops every six weeks and intermediate iterations between. The cadence matches Anthropic's release-train pattern and is materially faster than Google's Gemma cadence, which puts Meta as the open-weights pace-setter for the first time since the Llama 3.1 release in mid-2024.
Meta — Llama 5.1 405B release → · HuggingFace — Llama 5.1 405B model card → · Meta AI Blog — Six-week Llama cadence →