Skip to content

README.md

Latest commit

 

History

History
194 lines (129 loc) · 7.78 KB

README.md

File metadata and controls

194 lines (129 loc) · 7.78 KB
Raw

GLM-5.3-Flash API Python Quickstart with CometAPI

Runnable Python examples for calling GLM-5.3-Flash through CometAPI's OpenAI-compatible API: chat completions, streaming, multimodal input, reasoning controls, and temporary launch pricing.

Quick answer: use the model ID glm-5.3-flash, the CometAPI base URL https://api.cometapi.com/v1, and the Chat Completions endpoint POST /chat/completions. Check the live CometAPI model details before shipping because model availability, fields, and promotional pricing can change.

GLM-5.3-Flash at a glance

GLM-5.3-Flash is the first natively multimodal model in the GLM-5 series, according to Z.ai's launch announcement. It is positioned for coding agents, long-context work, and visual feedback loops while keeping inference costs low.

Setting Value
API model ID glm-5.3-flash
Provider Z.ai
API style OpenAI-compatible Chat Completions
CometAPI base URL https://api.cometapi.com/v1
Endpoint POST /chat/completions
Modalities described by the provider Text, image, and video input
Context window Up to 1M tokens, according to the launch materials
Reasoning Thinking is enabled; use the live route's supported effort values

The model is different from the text-focused glm-5.3. Do not silently substitute one model ID for the other in production configuration.

Promotional pricing snapshot

The CometAPI update shown on August 27, 2026 describes a limited-time 50% discount for glm-5.3-flash. Treat the discount as a dated launch promotion rather than a permanent budget assumption.

Price per 1M tokens Standard reference 50% promotion
New input $0.15 $0.075
Cached input $0.03 $0.015
Output $0.50 $0.25

The promotion is shown as ending at 24:00 on September 9, 2026 (UTC+8, Singapore Time), equivalent to 16:00 UTC on September 9, 2026. Re-check the live CometAPI model details and billing page before using these numbers in a quote or cost model.

What is in this repository?

  • examples/quickstart.py — the smallest working text request
  • examples/streaming.py — prints text deltas as they arrive
  • examples/multimodal.py — sends text plus an image URL
  • examples/reasoning_controls.py — shows explicit thinking settings
  • examples/promo_cost.py — estimates token cost at standard and promotional rates
  • .env.example — environment-variable template with no real credentials

Quickstart

1. Install the dependency

python -m venv .venv

# macOS / Linux
source .venv/bin/activate

# Windows PowerShell
.venv\Scripts\Activate.ps1

pip install -r requirements.txt

2. Set your API key

Copy .env.example to your local environment and export the variables. The examples read environment variables directly so a secret is never committed to Git.

# macOS / Linux
export COMETAPI_API_KEY="YOUR_COMETAPI_KEY"
export DEEPSEEK_MODEL="glm-5.3-flash"

# Windows PowerShell
$env:COMETAPI_API_KEY = "YOUR_COMETAPI_KEY"
$env:GLM_MODEL = "glm-5.3-flash"

The DEEPSEEK_MODEL line is optional and is retained only for compatibility with the earlier quickstart pattern. The GLM examples use GLM_MODEL.

3. Run the first request

python examples/quickstart.py

The core call is:

import os

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["COMETAPI_API_KEY"],
    base_url="https://api.cometapi.com/v1",
)

response = client.chat.completions.create(
    model="glm-5.3-flash",
    messages=[
        {"role": "user", "content": "Give me one practical Python testing tip."}
    ],
)

print(response.choices[0].message.content)

Streaming responses

Streaming is useful for agent interfaces and long answers because the UI can render text before the full response is complete:

python examples/streaming.py

The example uses the standard OpenAI Python SDK shape with stream=True. Add timeouts, retry policy, request IDs, and structured logging before using it in a production worker.

Multimodal input

The GLM-5.3-Flash launch materials describe native image and video input. The example sends an image URL using the OpenAI-compatible content-block format:

python examples/multimodal.py

Set IMAGE_URL to a test image you are allowed to access. A specific CometAPI route may expose a subset of the provider's modalities, so verify the live model details and error behavior before depending on image or video input.

Thinking and reasoning controls

For repeatable evaluations, set the reasoning fields explicitly when the live CometAPI route supports them:

python examples/reasoning_controls.py

The example sends:

extra_body={
    "thinking": {"type": "enabled"},
    "reasoning_effort": "max",
}

Z.ai's launch materials describe enabled thinking for GLM-5.3-Flash. If the route rejects a provider-specific field, remove that field and follow the current CometAPI API reference rather than guessing at a fallback.

Cost estimation

Use the included helper to compare the temporary promotion with the standard reference price:

python examples/promo_cost.py --input 1000000 --cached-input 200000 --output 300000

The helper is an estimate only. Actual billing depends on the live CometAPI rate card, tokenization, caching rules, and the date of the request. Budget against the standard rate after the promotion ends.

Production checklist

  1. Keep COMETAPI_API_KEY in a secret manager or environment variable.
  2. Pin glm-5.3-flash in configuration and validate allowed model IDs.
  3. Set network timeouts and bounded retries for POST /chat/completions.
  4. Log latency, input/output tokens, retry count, and provider request IDs without logging secrets.
  5. Test text-only and multimodal prompts separately; do not assume every route supports every input type.
  6. Compare the standard rate with the temporary promotion before approving a budget.
  7. Re-check the live CometAPI model details after September 9, 2026.

FAQ

What is the GLM-5.3-Flash API model ID?

The model ID used by this repository is glm-5.3-flash. It is not the same as glm-5.3.

How do I call GLM-5.3-Flash with CometAPI?

Install the OpenAI Python package, set COMETAPI_API_KEY, use https://api.cometapi.com/v1 as base_url, and call client.chat.completions.create(model="glm-5.3-flash", ...).

Is GLM-5.3-Flash multimodal?

Z.ai describes it as natively multimodal with text, image, and video input. The exact modalities exposed by a hosted API route should be verified in the current CometAPI model details.

How long does the 50% promotion last?

The update snapshot says the promotion ends at 24:00 UTC+8 on September 9, 2026, which is 16:00 UTC. Treat this as temporary and verify the live rate card.

Does this repository include model weights?

No. It contains small API client examples only. For local deployment information, read the official Z.ai announcement and model repository linked below.

Source and verification links

This is an original, code-focused developer adaptation of the linked documentation. Provider model IDs, capabilities, availability, and pricing can change; verify the live sources before shipping.