Zhipu Z.ai 7/13 GLM-4.6: AutoGLM-Coder refresh + agent scaffold, commercial-only on BigModel

2 sources

Zhipu Z.ai on 2026-07-13 publishes GLM-4.6 with AutoGLM-Coder coding-specialist post-training + agent scaffold. Self-reported figures include MMLU-Pro 81.2, GPQA-Diamond 73.5, HumanEval+ 90.8, MultiPL-E 87.6, SWE-Bench Verified 67.8, Terminal-Bench 2.1 74.6. License: commercial-only via BigModel / Z.ai; open-source variants stated as a future commitment, not a current artifact.

Why now

It lands one week after Zhipu's GLM-5.2 entry on 07-09 — adding a 7/13 GLM-4.6 update to the cluster, on the same day as five other vendor-aligned drops. For a Chinese-vendor creator, this is one of two commercial-only Chinese updates today (with ByteDance Doubao 1.6 Pro).

Why it is worth publishing

Best for Chinese-vendor creators + AutoGLM / agent-scaffold creators. The 'open-source variants will follow separately' is a vendor-stated future commitment and a useful legal hook; the AutoGLM-Coder framing is the agentic hook.

Evidence basis

Refresh of GLM-4 family + AutoGLM-Coder post-training + commercial-only BigModel license + future open-source commitment. Quad heat, no media corroboration today.

“Zhipu just refreshed GLM-4 — AutoGLM-Coder post-training, agent scaffold, and 'open source will follow' is the future bit.”

Angle

Frame GLM-4.6 as a refresh of the GLM-4 family (NOT a base-model swap) with AutoGLM-Coder coding-specialist post-training — and explicitly note that the open-source variant is a vendor-stated future commitment, not a current artifact. Lead with the AutoGLM-Coder framing, then the agent scaffold, then the commercial-only license framing.

Format

Carousel + a single-card licensing callout

Demo idea

Carousel card 1: 'GLM-4.6 is a GLM-4 refresh, not a base-model swap'; card 2: 'AutoGLM-Coder post-training + agent scaffold'; card 3: 'HumanEval+ 90.8 / SWE-Bench Verified 67.8 / Terminal-Bench 2.1 74.6 — vendor-supplied'; card 4: 'commercial-only via BigModel / Z.ai today'; card 5: 'open-source variant is a stated future commitment, not a current artifact — re-verify before quoting'. Pairs with the ByteDance Doubao 1.6 Pro card.

Platform notes

Refresh vs base-model swap (medium risk): do not call GLM-4.6 'a new base model' or 'GLM-5.' When viewers ask 'is GLM-4.6 better than GLM-4 prior at coding,' pin the link to the Zhipu platform news and quote only the deltas that are stated. The 'open-source variants will follow separately' framing is a vendor-stated future commitment — re-verify when the artifact ships.

Usable claims

Zhipu Z.ai on 2026-07-13 published a GLM-4.6 refresh with AutoGLM-Coder coding-specialist post-training + agent scaffold; self-reported figures include MMLU-Pro 81.2, GPQA-Diamond 73.5, HumanEval+ 90.8, MultiPL-E 87.6, SWE-Bench Verified 67.8, Terminal-Bench 2.1 74.6; commercial-only via BigModel / Z.ai; open-source variants will follow separately.

Evidence pipeline

From the news

Zhipu Z.ai 7/13 GLM-4.6: AutoGLM-Coder refresh + agent scaffold, commercial-only on BigModel

Breakdown

Walks GLM-4.6 as a refresh (not a base-model swap) of the GLM-4 family, paired with AutoGLM-Coder coding-specialist post-training + agent scaffold, the HumanEval+ 90.8 / MultiPL-E 87.6 / SWE-Bench Verified 67.8 / Terminal-Bench 2.1 74.6 benchmark headlines, the commercial-only BigModel license, and the vendor-stated future commitment of an open-source variant — explicitly flagged as future, not a current artifact.

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 with ByteDance Doubao 1.6 Pro — show as 'two Chinese-vendor commercial-only updates today,' with columns for license type + benchmark + vendor-stated future commitment.
If you cover GLM-4 family timelines, update the existing 'GLM-5.2 mid-July' entry to slot GLM-4.6 in as a refresh — do not duplicate the GLM-5.2 entry.
Use as the cluster's 'AutoGLM-Coder post-training' hook: 'today's cluster also drops Zhipu's AutoGLM-Coder refresh — pair with OpenAI / Anthropic / Google coding benchmarks if you ship a single-day coding-creator thread.'