Describe the feature or problem you'd like to solve
Add support for selecting OpenAI's Flex tier (service_tier: flex) in the Copilot CLI model selector and configuration, allowing users to cut token usage costs by 50% for non-time-critical tasks.
Proposed solution
Motivation
OpenAI Flex tier costs 50% less than Standard tier and according to monitoring data from OpenRouter has roughly 30% higher E2E latency (e.g. 9.02s vs 11.56s for gpt-5.6-sol):
| Model |
Output Price (Standard) |
Output Price (Flex) |
E2E Latency (Standard) |
E2E Latency (Flex) |
gpt-6-astra |
$50 |
$25 |
10.95s |
11.34s |
gpt-5.6-sol |
$20 |
$10 |
9.02s |
11.56s |
For many agentic tasks small increases of latency do not significantly impact user experience.
Additionally, E2E latency for many agentic tasks depends more on tool call durations than on LLM response times (e.g. large C++ or Gradle builds, running extensive test suits, etc). Using Flex tier for these kinds of tasks would reduce cost without significantly impacting task completion time.
Therefore, offering Flex tier would allow some developers to perform roughly twice as many agentic tasks withing a given AI budget.
Proposed Solution
- Interactive Model Picker: When selecting an OpenAI model, present a tier selection prompt (Standard vs. Flex) alongside the reasoning effort selection.
- CLI Flags: Add
--tier <standard|flex> (or --service-tier <standard|flex>) for scripted runs and aliases.
Example prompts or workflows
No response
Additional context
No response
Describe the feature or problem you'd like to solve
Add support for selecting OpenAI's Flex tier (
service_tier: flex) in the Copilot CLI model selector and configuration, allowing users to cut token usage costs by 50% for non-time-critical tasks.Proposed solution
Motivation
OpenAI Flex tier costs 50% less than Standard tier and according to monitoring data from OpenRouter has roughly 30% higher E2E latency (e.g. 9.02s vs 11.56s for
gpt-5.6-sol):gpt-6-astragpt-5.6-solFor many agentic tasks small increases of latency do not significantly impact user experience.
Additionally, E2E latency for many agentic tasks depends more on tool call durations than on LLM response times (e.g. large C++ or Gradle builds, running extensive test suits, etc). Using Flex tier for these kinds of tasks would reduce cost without significantly impacting task completion time.
Therefore, offering Flex tier would allow some developers to perform roughly twice as many agentic tasks withing a given AI budget.
Proposed Solution
--tier <standard|flex>(or--service-tier <standard|flex>) for scripted runs and aliases.Example prompts or workflows
No response
Additional context
No response