Flex Tier

Flex trades waiting time for price. The models, context length and output quality are identical to the standard tier — the only difference is a lower queue priority, and in return you pay 40% less.

This site is not an official OpenAI, Anthropic or xAI service, and has no affiliation, authorization or partnership with those companies.

Pricing

Per 1M tokens (input / output):

Model	Standard	Flex
gpt-5.5	$10 / $60	$6 / $36
gpt-5.6-terra	$4 / $24	$2.4 / $14.4
gpt-5.4-mini	$1.5 / $9	$0.9 / $5.4
gpt-5.6-luna	$0.4 / $2.4	$0.24 / $1.44

Every model is 40% cheaper.

Trade-offs

• Requests queue. Latency is unpredictable: the same request may return in seconds or take several minutes.
• When upstream load is high you will get a 429 and have to retry yourself.
• Set your client timeout to 300 seconds or more, otherwise queuing looks like an outage.

When to use it

Good fit: batch processing, offline analysis, document translation, code review, data cleaning — anything where you are not watching the screen.

Poor fit: Claude Code, IDE plugins, chat UIs and other interactive workflows. Keep those on the standard tier or they will feel sluggish.

How to use it

1. Open the Tokens page in the console
2. Create a token and set its group to Flex
3. Call the API with that token's key — model names stay the same

curl https://api.treenew.online/v1/chat/completions \
  -H "Authorization: Bearer <your Flex token key>" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5.5",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Keep two tokens

Create one token per tier and pick by workload: standard for interactive work, Flex for batch jobs.

FAQ

Is Flex output lower quality?

No. Same model, same parameters — only the queue priority differs.

How do I confirm a request actually used Flex?

The response contains "service_tier": "flex". Standard-tier responses show "service_tier": "default".

How long is the queue?

No guarantees. When traffic is light it is close to the standard tier; at peak it can be noticeably slower or return 429. If your task is time-sensitive, use the standard tier.