[Bug]: Add baseten/zai-org/GLM-5.3 to the model pricing registry #39342

Check for existing issues

• I searched the existing LiteLLM issues and pull requests for baseten/zai-org/GLM-5.3; no matching report or change exists.

What happened?

LiteLLM's current model pricing registry does not contain baseten/zai-org/GLM-5.3 in either model_prices_and_context_window.json or litellm/model_prices_and_context_window_backup.json. A gateway using LiteLLM's native Baseten adapter can route requests to baseten/zai-org/GLM-5.3, but LiteLLM reports that the model is not mapped and cannot supply its provider pricing, token limits, or capability metadata unless every deployment duplicates that information locally.

The existing zai/glm-5.3 entry is not a fallback for this route: it belongs to the Z.AI provider and has different cached-input pricing. The existing Baseten registry only includes baseten/zai-org/GLM-5.

Expected behavior: the built-in registry contains a baseten/zai-org/GLM-5.3 entry in both generated registry files, backed by Baseten's public Model API documentation and pricing. Current Baseten values are $1.40 per million input tokens, $0.14 per million cached input tokens, and $4.40 per million output tokens; its live Model API advertises a 1,048,576-token context and a 262,144-token maximum output.

Suggested metadata includes the Baseten provider, chat mode, the three price fields, token limits, always-on reasoning with low/high/max, function/tool calling, parallel function calling, response schemas, prompt caching, and text-only modality. Please update both registry JSON files and add the normal metadata regression test so they cannot drift.

Baseten sources:

• https://www.baseten.co/pricing/
• https://docs.baseten.co/inference/model-apis/overview
• https://docs.baseten.co/inference/model-apis/reasoning

User Flow

Before a hypothetical fix: a LiteLLM Proxy administrator exposes Baseten GLM-5.3, but the gateway cannot obtain the model's built-in pricing and limits.

1. The administrator configures a Proxy deployment with model: baseten/zai-org/GLM-5.3 and a Baseten API key.
2. A user sends POST http://localhost:4000/v1/chat/completions with model zai-org/GLM-5.3.
3. The Baseten request can be routed, but the gateway has no built-in baseten/zai-org/GLM-5.3 model metadata; inspecting the model through LiteLLM reports This model isn't mapped yet unless the administrator duplicates prices and limits in the deployment configuration.
4. The administrator must maintain a local pricing override, which can drift from Baseten's published rates.

After a hypothetical fix: the same Baseten route resolves its provider metadata from LiteLLM's built-in registry.

1. The administrator configures the same Proxy deployment with model: baseten/zai-org/GLM-5.3 and a Baseten API key.
2. A user sends the same POST http://localhost:4000/v1/chat/completions request with model zai-org/GLM-5.3.
3. LiteLLM resolves the Baseten GLM-5.3 pricing, token limits, and capabilities without a deployment-local pricing override.
4. The administrator can remove the temporary local metadata and receive registry updates through normal LiteLLM upgrades.

Proof the bug occurs

Config / setup:

model_list:
  - model_name: zai-org/GLM-5.3
    litellm_params:
      model: baseten/zai-org/GLM-5.3
      api_key: os.environ/BASETEN_API_KEY

Versions checked:

• LiteLLM v1.99.0, commit fa647f742d7baefe8eb1181899d9c81b41559772
• Current litellm_internal_staging, commit 2ce4e3f8a99e12efce9433640059d9fca7bfb448

Exact v1.99.0 runtime reproduction using the immutable published image:

$ docker run --rm -i --entrypoint python ghcr.io/berriai/litellm:1.99.0@sha256:570a872d2fde8f1bc4a147634810941103c14697270f0f2918c5aa7d8201cac5 - <<'PY'
import json
import litellm

print("model key present:", "baseten/zai-org/GLM-5.3" in litellm.model_cost)
try:
    info = litellm.get_model_info(model="zai-org/GLM-5.3", custom_llm_provider="baseten")
    print(json.dumps(info, indent=2, sort_keys=True, default=str))
except Exception as exc:
    print(f"{type(exc).__name__}: {exc}")
PY
model key present: False
Exception: This model isn't mapped yet. model=zai-org/GLM-5.3, custom_llm_provider=baseten. Add it here - https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json.

Current default-branch registry reproduction:

$ sha=2ce4e3f8a99e12efce9433640059d9fca7bfb448
$ for path in model_prices_and_context_window.json litellm/model_prices_and_context_window_backup.json; do curl -fsSL "https://raw.githubusercontent.com/BerriAI/litellm/$sha/$path" | jq '."baseten/zai-org/GLM-5.3"'; done
null
null

This is a static model-registry defect and does not require sending a billable provider request: the missing exact key is the failing boundary. Baseten's public sources above provide the expected live prices and limits.

What part of LiteLLM is this about?

Other — model pricing and context-window registry.

What LiteLLM version are you on?

v1.99.0; also reproduced at current litellm_internal_staging commit 2ce4e3f8a99e12efce9433640059d9fca7bfb448.