GLM-5.3

GLM-5.3 is a large-scale reasoning model built for complex software engineering and long-horizon agent tasks. It supports text input and output with a 1,048,576-token context window.

At a glance

Key capabilities

Quickstart

from sarvamai import SarvamAI
client = SarvamAI(api_subscription_key="YOUR_SARVAM_API_KEY")
response = client.chat.completions_v2(
    model="glm5.3",
    messages=[
        {
            "role": "user",
            "content": "Explain the difference between a B-tree and a B+ tree index.",
        }
    ],
    temperature=1,
    top_p=0.95,
    max_tokens=8000,
)
print(response.choices[0].message.content)

Supported parameters

model: "glm5.3" on Chat Completion V2 and Responses. Full request/response schemas live on those API reference pages; this table covers what’s actually accepted for this model.

All 400s above are invalid_request_error for unsupported/invalid parameters.

Reasoning

GLM-5.3 has three reasoning modes, low, high, and max, but which values you can pass depends on the API:

Read the reasoning trace from choices[].message.reasoning_content on chat, or from the reasoning item’s content[0].text in output on Responses. Its summary is empty unless you set reasoning.summary to "auto" or "detailed". Reasoning tokens are billed as output tokens either way, and count against max_tokens on chat (defaults to 2048 when omitted) or max_output_tokens on Responses (no default, omitting it means unbounded generation). At the default max effort, reasoning often uses several thousand tokens, so leave room for the answer.

Reasoning can’t be turned off, but at low effort GLM-5.3 often skips it on simple prompts: reasoning_tokens is 0 and Responses returns no reasoning item.

Response fields (chat)

Full response shape: Chat Completion V2.

Known limitations

• Output token budget. max_tokens on chat defaults to 2048 when omitted; max_output_tokens on Responses has no default and allows unbounded generation. Reasoning counts against whichever budget applies, so a small max_tokens on chat can leave little or no room for the actual answer.
• Context-length checks are approximate, and more conservative for Latin-script text than for Indic scripts. Treat the model’s own limit as authoritative.

Parameter errors (n, logprobs, stop, and similar) are documented on Chat Completion V2.

Limits and errors