[quickbeard]

What

In packages/codev-gateway/src/model-limits.ts:

• adds zai-org/GLM-5.3-Flash at 262,144 tokens, the same window as MiniMax/MiniMax-M3
• removes zai-org/GLM-4.7-cc

The table now carries only the models whose window differs from the default.

Why

Without a table entry GLM-5.3-Flash resolved to DEFAULT_LIMITS (the safe 200K fallback), so every config written for it — TUI /configure, codev configure, the desktop SSO sign-in, and the sandbox guest block — understated the window in the "% context used" gauge and started auto-compacting about 62K tokens earlier than the model can actually take.

The GLM-4.7-cc entry's 200000 restated DEFAULT_WINDOW exactly, so removing it changes nothing: limitsFor returns the same { context: 200000, trigger: 180000 } through the default tier, and a config block for that model is written byte-identically. The file's own comment already said an entry restating the default is one more thing to keep in sync.

Numbers

limitsFor("zai-org/GLM-5.3-Flash") → { context: 262144, trigger: 235930 } (the uniform 90%). As with M3, trigger + COMPACT_RESERVED (235930 + 40000) exceeds the window, so declaredInput's clamp pins limit.input at 262144 and the effective trigger is 262144 − 40000 = 222144 (~85% of the window). That clamp is pre-existing behaviour for every model with a window under 400K, not something this PR introduces.

	context	limit.input	effective trigger
GLM-5.3-Flash before (200K default)	200 000	200 000	160 000
GLM-5.3-Flash after	262 144	262 144	222 144
GLM-4.7-cc, before and after	200 000	200 000	160 000

Assumption to confirm

The 262,144 window is taken from "configure it like MiniMax-M3" — M3 is capped there by the deployment's resources rather than by the model. If the gateway actually serves GLM-5.3-Flash with a different window, say so and I'll change the one number. Guessing too high is the direction that hurts (a session can run past the real ceiling and 400 mid-chat), so this is worth a second pair of eyes. It can also be overridden without a release: the moment an admin sets max_input_tokens on the model in LiteLLM, fetchModelWindows caches it and the remote tier outranks this table.

Scope

• Existing configs are not rewritten — a user picks the new numbers up by re-running a configure flow (or the hub).
• No DIVERGENCES.md row: packages/codev-gateway is a fork-only package, and no upstream file is touched.
• The parallel table in codev-hub gets both changes in feat(model-limits): table the GLM-5.3-Flash window, drop GLM-4.7-cc codev-hub#262 — the two are separate copies by design and have to move together.

Testing

• bun test in packages/codev-gateway — 65 pass, 0 fail
• bun typecheck in packages/codev-gateway — clean

🤖 Generated with Claude Code