Shareable benchmark table

Frozen GLM-5.3 + ZCode agent-system lock

The agent system under test, recorded in agent.lock.json before any run and hashed into every run ledger.

Table 17 · Agent-system lock

Frozen GLM-5.3 + ZCode agent-system lock

Agent system · GLM-5.3 + ZCode CLIn = 1 run / cellOperator contamination disclosedDeterministic evaluatorNot a model leaderboard
The agent system under test, recorded in agent.lock.json before any run and hashed into every run ledger.
FieldValue
Model identifierbuiltin:zai-start-plan/GLM-5.3
Runtime engineZCode CLI/desktop harness (one stateful session)
ZCode app3.8.1 · CLI bundle sha256 9318f60fb8c2c3bc…
Providerzai (OAuth), Anthropic-compatible endpoint
Reported limitscontext 1,000,000 · output 128,000
Inference controlnone (no seed, no sampler control)
Token meteringnot available — recorded n/a, never zero
SandboxDocker --network none · cap-drop ALL · no-new-privileges · solc 0.8.35
HostApple M4 Pro · 24 GB · arm64 · macOS 27.0
Lock frozenbefore first run · 2026-08-21T19:59:41Z
Replicates1 per cell (deviation from parent 6/cell justified pre-run)
Table 17The agent system under test, recorded in agent.lock.json before any run and hashed into every run ledger.

Inference is cloud-served through the ZCode runtime: sampler settings are not operator-controlled and token telemetry does not exist (recorded n/a, never imputed). The arm deviates from the parent 6-replicate plan: 1 run per cell, justified before collection because a single stateful session cannot produce independent replicates.

BMSbms.kyokasuigetsu.xyz
Static image: https://bms.kyokasuigetsu.xyz/tables/glm-agent-lock.png