What It Does
Flex Processing is an OpenAI service tier that runs a request at 50% of
standard pricing in exchange for slower responses and occasional resource
unavailability. CBorg exposes it as a set of model aliases: append -flex to
any eligible OpenAI model name, and the request is served on OpenAI’s flex tier
and billed against your API budget at the discounted rate automatically.
Flex is best suited to workloads that tolerate latency – batch jobs, background processing, evaluations, bulk generation. It is not recommended for interactive chat, where the added latency is noticeable and requests may occasionally be declined under load.
Note
Higher latency, possible declines: On the flex tier OpenAI may take
significantly longer to respond, and under heavy load may return a
429 resource_unavailable error instead of serving the request. Use flex for
work that can wait or retry, and keep standard (non--flex) models for
interactive use.
How to Use
Append -flex to the model name. The suffix sits after the reasoning-effort
segment and before any compression segment:
<model>[-<reasoning-effort>]-flex[-compact|-compressed]
Examples:
gpt-5-mini-flexgpt-5.6-sol-flexgpt-5.6-sol-xhigh-flex(flex +xhighreasoning effort)gpt-5.6-sol-xhigh-flex-compressed(flex + reasoning + Headroom compression)
curl https://api.cborg.lbl.gov/chat/completions \
-H "Authorization: Bearer $CBORG_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-5-mini-flex",
"messages": [{"role": "user", "content": "Summarize this batch of documents..."}]
}'
With the OpenAI Python SDK, use the -flex model name directly:
import os
import openai
client = openai.OpenAI(
api_key=os.environ["CBORG_API_KEY"],
base_url="https://api.cborg.lbl.gov",
)
response = client.chat.completions.create(
model="gpt-5-mini-flex",
messages=[{"role": "user", "content": "Summarize this batch of documents..."}],
)
Pricing
Flex requests are billed at exactly 50% of the standard input and output token rates for the same model. Spend and budget tracking reflect the discount automatically – no additional configuration is required, and the reduced cost is recorded per request. Cached-input and reasoning tokens are discounted on the same basis.
For the exact per-token rate of any model, see the AI Models page; halve the listed input/output prices to get the flex rate.
Eligible Models
Flex is an OpenAI feature and applies only to directly OpenAI-sourced models that
OpenAI offers the flex tier for. It is model-specific, not a blanket GPT-5
feature. Eligible models on CBorg (each also available in its reasoning-effort
variants, e.g. -high, -xhigh):
gpt-5,gpt-5-mini,gpt-5-nanogpt-5.4,gpt-5.4-mini,gpt-5.4-nano,gpt-5.4-progpt-5.5,gpt-5.5-progpt-5.6-sol,gpt-5.6-terra,gpt-5.6-luna
Models without a -flex variant (including the GPT-5.1, GPT-5.2, and Codex
families, and non-OpenAI models served through CBorg) do not support flex
processing.
See Also
- AI Models – full model list and per-token pricing
- Headroom Compression – combine with
-flexfor further token savings