feat(model_prices): add zai/glm-5.2 vendor entry

[wait4xx]

Summary

Adds the official Z.AI vendor entry zai/glm-5.2 to model_prices_and_context_window.json and litellm/model_prices_and_context_window_backup.json.

The registry currently covers GLM-5.2 only through third-party deployments (dashscope/glm-5.2, together_ai/zai-org/GLM-5.2, deepinfra/zai-org/glm-5.2, cloudflare/@cf/zai-org/glm-5.2, novita/zai-org/glm-5.2, ...). The two-segment zai/ vendor entry is missing, so zai/glm-5.2 resolves without pricing/context metadata.

Values (per official docs)

Field	Value	Source
input_cost_per_token	1.4e-06 ($1.4/M)	pricing page, GLM-5.2 row
cache_read_input_token_cost	2.6e-07 ($0.26/M)	pricing page, "Cached Input"
cache_creation_input_token_cost	0	"Cached Input Storage: Limited-time Free"
output_cost_per_token	4.4e-06 ($4.4/M)	pricing page, GLM-5.2 row
max_input_tokens	1000000	model page, "Context Length: 1M"
max_output_tokens	128000	model page, "Maximum Output Tokens: 128K"
capability flags	function calling / prompt caching / reasoning / tool choice = true	model page capability cards

Sources:

• https://docs.z.ai/guides/llm/glm-5.2
• https://docs.z.ai/guides/overview/pricing

The entry mirrors sibling zai/glm-5.3 field-for-field: on the official pricing page GLM-5.2 and GLM-5.3 sit in the same row group ($1.4 / $0.26 / $4.4), and the model pages document the same 1M / 128K limits and the same capability set (text-only input/output, so no supports_vision, consistent with the sibling).

Tests

• Both JSONs parse after the change
• New blocks in the two files are byte-identical; git diff shows only the two +15-line additions
• No duplicate zai/glm-5.2 key; no other registry entries touched

[CLAassistant]

All committers have signed the CLA.

[greptile-apps]

Greptile Summary

Adds direct Z.AI metadata for zai/glm-5.2 to both synchronized model registries.

• Records input, output, and cached-input pricing.
• Defines the model’s input and output token limits.
• Advertises function calling, prompt caching, reasoning, and tool-choice support.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
model_prices_and_context_window.json	Adds the direct zai/glm-5.2 pricing, context-limit, provider, and capability metadata.
litellm/model_prices_and_context_window_backup.json	Mirrors the new zai/glm-5.2 entry from the primary model registry.

_{Reviews (2): Last reviewed commit: "feat(model_prices): add zai/glm-5.2 vend..." | Re-trigger Greptile}

[codspeed-hq]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing wait4xx:fix/zai-glm-5.2-vendor-entry (f2b695a) with litellm_internal_staging (d44d281)}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[wait4xx]

Note on the proxy-infra / Run tests failure: it looks pre-existing on the base branch rather than caused by this change.

The failing test (test_savings.py::test_a_baseline_with_no_cache_read_rate_is_charged_its_input_rate) asserts that xai/grok-4 has no cache_read_input_token_cost ("pick a baseline with no cache-read rate"). On the PR base (litellm_internal_staging @ d44d281, without this PR's commit) the registry entry already carries it:

"xai/grok-4": { ..., "cache_read_input_token_cost": 2e-07, ... }

The entry is byte-identical on this PR's branch — the diff only adds the zai/glm-5.2 key. The same test passes on main (10631eb), where xai/grok-4 does not carry the field yet, so the regression came in via an earlier staging commit, not this one.

Everything that validates the registry change itself is green on this PR: validate-model-prices-json, Greptile 5/5, CodSpeed no-change.

[devin-ai-integration]

Superseded by the rolling registry PR #38990, which absorbs this change (values re-verified against the official provider docs). Closing per the one-rolling-registry-PR policy.