OpenAI 7/13 gpt-oss refresh: open-weight Apache-2.0 + OpenAI usage-policy addendum

3 sources

OpenAI on 2026-07-13 publishes a refresh of the gpt-oss open-weight line — 120B MoE and 20B dense variants with new post-training snapshots. Self-reported benchmark headlines include MMLU-Pro 84.6, GPQA-Diamond 76.2, SWE-Bench Verified 68.3, BFCL v3 65.4, Terminal-Bench 2.1 76.8. License remains Apache-2.0 with the OpenAI usage-policy addendum.

Why now

It lands on the same day as five other vendor-aligned model drops, and it's the only US frontier-lab open-weight move in the cluster — creators can carry 'the US side of the open-weight day' as the story.

Why it is worth publishing

Best for open-weight creators + Apache-2.0 distribution creators. The OpenAI usage-policy addendum is the legal hook; the BFCL v3 framing is the agentic hook.

Evidence basis

OpenAI open-weight refresh + Apache-2.0 + first Decoder item of the day (Independently verified badge). Triple-stack heat.

“OpenAI just refreshed its only open-weight line — and kept it on Apache-2.0, with one new addendum and four new digits to argue about.”

Angle

Frame gpt-oss refresh as the 'OpenAI keeps Apache-2.0' open-weight story — not as 'gpt-oss catches GPT 5.5 on benchmarks.' Lead with the badge moment (Independently verified), then the architecture (120B MoE + 20B dense), then the addendum (OpenAI usage-policy) — that ordering keeps the legal framing visible without overclaiming.

Format

Carousel + 30-second talking-head follow-up

Demo idea

Carousel card 1: title; card 2: '120B MoE + 20B dense' (one line each); card 3: 'MMLU-Pro 84.6 / SWE-Bench Verified 68.3 / BFCL v3 65.4' with the qualifier 'vendor-supplied, not reproduced'; card 4: 'Independently verified' badge callout; card 5: 'Apache-2.0 + the OpenAI usage-policy addendum'. Then a 30-second talking-head: 'here's what the OpenAI usage-policy addendum overlays on Apache-2.0 — quote only what the page states; don't generalize'.

Platform notes

Self-reported benchmarks (medium risk): the OpenAI usage-policy addendum boundary was not extracted in the captured summary — quote only what the OpenAI index page and the HF card state. The Decoder item is a paraphrase, not an independent reproduction of any specific score. When viewers ask 'is gpt-oss 120B catching up to GPT 5.5,' don't extrapolate from a single vendor chart, especially across a 16x activated-parameter gap and a different post-training pipeline.

Usable claims

OpenAI on 2026-07-13 published a refresh of the gpt-oss open-weight line (120B MoE + 20B dense) with new post-training snapshots; self-reported benchmark numbers include MMLU-Pro 84.6, GPQA-Diamond 76.2, LiveCodeBench v6 72.1, SWE-Bench Verified 68.3, BFCL v3 65.4, Terminal-Bench 2.1 76.8; license remains Apache-2.0 with the OpenAI usage-policy addendum.

Evidence pipeline

From the news

OpenAI 7/13 gpt-oss refresh: 120B MoE + 20B dense post-training snapshot under Apache-2.0

Breakdown

Walks the gpt-oss refresh's architecture (120B MoE + 20B dense), benchmark headlines (MMLU-Pro 84.6 / SWE-Bench Verified 68.3 / BFCL v3 65.4 / Terminal-Bench 2.1 76.8), and the 'Independently verified' badge moment paired with the Apache-2.0 + OpenAI usage-policy addendum framing. The article names the addendum but does not state its scope — addendum boundary text was not extracted in the captured summary.

Sources

Risks

Pin the link to each vendor page and the companion HF model card; quote only what those pages state; do not paraphrase self-reported benchmark numbers as third-party validation; for the Doubao topic specifically, pin the HF mirror card as a researcher-preview only and do not paraphrase the gating-to-API-only boundary.
Pin the link to each license text; quote only what the page states; do not paraphrase the baseline license as the sole operating condition; flag addenda by name without stating their scope; for Google Gemini 3.5 Pro, note that license lives with the API contract rather than a downloadable-weights license.

Demo ideas

Pair OpenAI gpt-oss with Meta Llama 4 Behemoth refresh (both US-lab refresh + both first Meta/OpenAI drop in the cluster) — show them as 'two US-lab weight drops on the same day,' one Apache-2.0 (gpt-oss) + one Llama 4 Community License (Llama 4).
Use the badge moment as the day's hook: 'today's batch has two Independently verified topics — OpenAI gpt-oss (tech-press) and Anthropic Sonnet 4.5 (tech-press) — read the addendum language carefully on both before quoting.'
For long-term open-weight tracking, build a 3-row Apache-2.0 timeline (DeepSeek-V4-Pro / gpt-oss refresh / future Mistral variant) — and pin the OpenAI usage-policy addendum as the legal column.