Cursor lists GLM 5.3 / Flash natively: 1M context, Other Models pool, from $1.4 / $0.15 per MTok

Core Background & Industry Pain Points

Z.ai’s open-weight GLM 5.3 / Flash shipped in August across many inference and agent hosts, but Cursor’s first-party picker long stopped at GLM 5.2—September forum threads still required a custom OpenAI-compatible base URL plus the exact glm-5.3 model ID. As of this verification, Cursor’s docs and Models & Pricing list GLM 5.3 and GLM 5.3 Flash as addable Z.ai models with dedicated pages—this is a product-surface inclusion, not another mirror rumor.

Architecture Highlights & Internals

Per the GLM 5.3 and GLM 5.3 Flash docs: open-weight models served via Cursor’s inference partners; full Cursor agent tool access; 1M-token context; reasoning levels Low / High / Max (High default). Flash adds image inputs. Billing uses the Other Models pool at API rates (Teams/Enterprise also incur Cursor Token Rate). Flat pricing across the full window—no long-context surcharge, no separate cache-write fee; GLM 5.3 matches GLM 5.2 rates. Enable via Settings → Models.

Authoritative Benchmarks & Measured Scores

Cursor does not publish Cursor-agent SWE/Terminal/CursorBench deltas for this listing—do not invent leaderboard claims. Verified $/MTok anchors from Cursor’s pricing table: GLM 5.3 $1.4 / $0.26 / $4.4 (input / cache read / output); GLM 5.3 Flash $0.15 / $0.029 / $0.5. Capability numbers still belong to Z.ai / third-party benches; “in the Cursor picker” ≠ a new weight release.

Developer Hands-on Guide

1. Update Cursor; Settings → Models → add GLM 5.3 or Flash (listed Hidden by default).
2. Need a plan with the Other Models pool (not India Start).
3. Prefer full 5.3 for hard agent loops; Flash for cheap / multimodal drafts.
4. BYOK via Z.ai Coding Plan URL remains a separate custom-model path.
5. Review subprocessors / residency for enterprise; do not conflate older vulnerability narratives with this listing unless a dated vendor advisory says so.