Cloud baseline

GLM-5.3 — 0.3317 on Gauntlet 7.0

glm-5.3 · scorer sb-7.0

Aug 24, 2026
Overall
0.3317
Excellent
no
Benchmark
Gauntlet 7.0
Recorded duration
103m 6s
Prompt tok
10.9M
Gen tok
365.5k

Token usage

Tokens: at least 10.9M in · 366k out · 10.7M cached (lower bound, harness session counters)

11,267,453 tokens over 128 model calls: every call of the run, the model's own and the harness's helper calls, counted from harness session counters. 2 calls have no token count, so these are a lower bound.

Per-response usage Z.ai returned, recorded request by request by the benchmark campaign and equal to the harness's session counters; Z.ai keeps no per-call bill to query. Reasoning is not reported separately. 2 requests started streaming but never reported usage (tokens unknown). Includes a first attempt that was cut and resumed (115,736 / 13,358 and 16,118 / 211 tokens).

  • glm-5.3

    In 10.9M · Out 366k · Cached 10.7M · 128 calls

Tier breakdown

A · structure & runtime1.00
B · behaviour0.96
C · vendor contract0.89
D · finesse0.80

Scoring detail

How this score was built

Per-check scorer output, as posted by the app. Expand a section to inspect its checks.

The number, exactly

(0.88 × 0.7406 core + 0.12 × 0.71 gate × 0.27 excellence) × 0.4920 critical = 0.3317

Critical defects compound a multiplier on the whole score (pre-severity 0.6742):

b_money_rendered 0.55 → ×0.82j_workflow_journey 0.00 → ×0.60

The core (88% of the total) is the weighted mean of the ten measured tiers. The last 12% is the excellence slice: it unlocks in proportion to the perfection conditions below (11 of 16 met here), then pays out at the excellence tier's own measured mean. Core 0.6517 + excellence 0.0225.

j_first_use 1.00j_workflow_journey 0.00j_error_state 0.30j_empty_state 1.00console_clean 0.00v_responsive_375 1.00v_dates_readable 1.00t_scene_binding 0.00x_conservation_residual 1.00r_no_row_loss 1.00p_drag_frames 0.00p_idle_flatness 1.00p_stream_apply 0.00p_under_stream 1.00p_api_latency 1.00p_sync_wall 1.00

Run notes and corrections

Cloud baseline — a single harness run session on the frozen sb-7 spec (Meridian Payments Console: two services, 12,288-payment collection, signed webhooks, maker/checker approval workflow, instanced WebGL 3D field), scored by the frozen sb-7 scorer. Screenshots are the scorer's own render-gate captures of the built app. Full sb-7 tier means (the schema's tierA–tierD carry only A–D): A 1.0000 · B 0.9584 · C 0.8922 · D 0.8000 · J 0.5786 · V 1.0000 · P 0.6667 · T 0.2057 · X 0.8333 · R 0.8819 · E 0.1877.

Token rates

Measured by the engine itself, one record per completed model call: prefill rate is prompt tokens over time-to-first-token, generation rate is completion tokens over the decode window. Medians per node.

NodeCallsPrompt tokGen tokPrefill tok/sGen tok/s
glm12810,901,940365,513——
fleet12810,901,940365,513——

Screenshots

Recorded views of the built application. Captions identify the available captures; the number of images does not imply a number of repair rounds.

1First render — before repairs
2Final render
33D field · 12,288 instanced columns · WebGL
4Approval workflow · drafts panel
5After sync

Run details

Model
glm-5.3

How the benchmark is scored