Identity and current terms
| Provider / API ID | OpenAI / gpt-6-astra |
|---|---|
| Context / maximum output | 1,050,000 / 128,000 tokens |
| Reasoning effort | Low, medium, high, xhigh, max |
| Standard short-context price | $10.00 uncached input, $1.00 cached input, $12.50 cache write, $50.00 output per 1M |
| Long-context condition | Above 272K input, 2× input/cache and 1.5× output apply to the full request |
When it belongs in the comparison
Astra is relevant when the job remains difficult after retrieval, tools, environment, and acceptance checks are under control: long-running multi-file work, ambiguous cross-layer failures, or workflows requiring broader end-to-end reasoning. This is a test-selection hypothesis, not a measured breakpoint.
Check organization access before designing the study around it. The model listing confirms the ID and terms; access can still depend on account and rollout conditions.
Same headline price does not mean same workflow cost
Astra and Fable 5.1 share a $10/$50 headline input/output pair, but their cache treatment, long-context pricing, integration, and harness conditions differ. A 300K-input request crosses Astra’s long-context price threshold; that fact should travel with any cost table.
Evidence limits
Independent coding-agent work can inform which comparison to run, but an Astra-with-Codex result versus Fable-with-Claude-Code includes system differences. The current research does not establish same-harness cost per accepted repository outcome.
Use the buyer guide to define the task and the cost analysis to preserve billing conditions.
Reader questions
Frequently asked questions
How much does GPT-6 Astra cost?
The standard short-context rates checked on 12 September 2026 are $10 per million uncached input tokens, $1 cached input, $12.50 cache write, and $50 output. Long-context and other service conditions can change them.
Is Astra proven better than Fable 5.1 for repository coding?
No same-task, same-harness AIModel.Reviews dataset supports that claim. Existing agent reports include harness differences and should be treated as external evidence.
Primary reading
Sources and update conditions
Rates, availability, model identity, and provider behavior can change. Recheck the linked sources before making a consequential purchase or sending repository data.