fix: [missing-model] tinfoil: glm-5-3

[github-actions]

Summary

Add missing Tinfoil model glm-5-3 (Z.ai GLM-5.3) as an override-only provider entry on zhipuai/glm-5.3, using Tinfoil’s live catalog pricing/context and documented always-on reasoning effort controls.

Changes

New file: providers/tinfoil/models/glm-5-3.toml

Field	Value
base_model	zhipuai/glm-5.3 (lab metadata already present)
reasoning_options	effort low, high, max (always-on; no none / no toggle)
cost.input	1.8 USD/MTok
cost.output	5.75 USD/MTok
cost.cache_read	0.45 USD/MTok
limit.context	1_048_576 (Tinfoil-served window; lab base is 1M)

Evidence

Claim	Source	What it establishes
Model exists as glm-5-3; pricing 1.8 / 5.75 / 0.45; context 1,048,576; reasoning + tool calling	https://inference.tinfoil.sh/v1/models	Live catalog entry for glm-5-3
Identity is GLM-5.3; 1M context; Z.ai/Inferact weights	https://docs.tinfoil.sh/models/chat.md	Docs card for glm-5-3 / GLM-5.3
Supported reasoning_effort: low, high, max; reasoning always on; enable_thinking not honoured	https://docs.tinfoil.sh/guides/reasoning.md	Per-model effort table + always-on note
Lab base already complete	models/zhipuai/glm-5.3.toml	base_model target exists

Validation

• Cross-checked the new entry against Tinfoil’s models API response, chat model docs, and reasoning guide.
• Matched style to existing Tinfoil peers (glm-5-2.toml, kimi-k3.toml, deepseek-v4-flash.toml).
• Did not run bun validate (no shell per instructions).

Review notes

• Host is a multi-model relay; effort set is Tinfoil’s documented surface for this model (low/high/max), not the broader GLM-5.2 Tinfoil set.
• Context override follows Tinfoil’s reported 1048576 (same pattern as deepseek-v4-flash); output limit inherits lab 131_072 (API does not publish a separate output max).
• None.

Closes #6151

Automated by the issue fixer: https://github.com/anomalyco/models.dev/actions/runs/33711238568