Gemini 4 Argon is on BazaarLink: 1M context, vision, tool calling
Google's flagship reasoning model Gemini 4 Argon is now available on BazaarLink through an OpenAI-compatible API. Pricing, features and a quick example.
Gemini 4 Argon, which Google announced on September 30, is now available on BazaarLink. The model ID is google/gemini-4-argon, and it works with your existing BazaarLink API key. No separate sign-up is needed.
What Gemini 4 Argon is
Argon is Google's current flagship reasoning model, built for software engineering, math and science problems, and agents that run many steps. Google reports 77.9% on DeepSWE v1.1 and 51.3% on AutomationBench (ranked first).
Specs
| Model ID | google/gemini-4-argon |
| Context length | 1,048,576 tokens |
| Max output per request | 65,536 tokens |
| Input | text, image, file, audio, video |
| Output | text |
| Reasoning | always on; set reasoning.effort to low / medium / high (default: medium) |
| Supports | tool calling (tools / tool_choice), structured output (response_format), streaming |
| Endpoints | /v1/chat/completions, /v1/responses, /v1/messages |
Pricing (USD per million tokens)
| Price | |
|---|---|
| Input | $2.00 |
| Output (including reasoning tokens) | $10.00 |
| Cache read | $0.10 |
This currently matches Google's introductory price (Google's listed standard rate is $4 input and $20 output).
Quick start
curl https://api.bazaarlink.ai/v1/chat/completions \
-H "Authorization: Bearer $BAZAARLINK_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "google/gemini-4-argon",
"messages": [{"role": "user", "content": "Check this code for race conditions"}],
"reasoning": {"effort": "medium"}
}'
With the OpenAI SDK, set base_url to https://api.bazaarlink.ai/v1 and model to google/gemini-4-argon. Tools that use the Anthropic format, such as Claude Code, can call /v1/messages.
Tips
- Reasoning is always on, so use
lowfor simple tasks. Reasoning tokens are billed at the output rate, and a lower effort saves a lot on easy prompts. - Don't set
max_tokenstoo low. Reasoning uses part of the output budget first, so a small limit can cut the reply off before the answer. - Use caching for long documents. Re-sending the same long context costs $0.10 per million tokens on a cache read.
- Set guardrails before an agent handles money or talks to outside parties. Give each agent its own key with a spending limit. See our earlier post, Gemini 4 Argon and Vending-Bench 2.
To compare prices and specs with other models, search for gemini-4-argon on the models page.
TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API