GPT-6 Astra vs GPT-5.6 Sol: Double the Input Price on Ofox

On Ofox, GPT-6 Astra costs twice as much per input token as GPT-5.6 Sol and 67% more per output token. $10.00 / $50.00 against $5.00 / $30.00. OpenAI shipped it on 3 September 2026, and the upgrade question is unusually simple to state: the price went up, so the output has to be worth more.

Choose a provider, then compare completed tasks

Prices checked on 16 September 2026 differ by provider. OpenAI’s current model comparison lists Astra at $10 input / $50 output and Sol at $4 / $20 per million tokens: 2.5× on both. The Ofox catalog lists Astra at $10 / $50 and Sol at $5 / $30: 2× input and about 1.67× output. Keep that distinction when interpreting the historical benchmark discussion below; neither ratio predicts a complete task’s bill.

Before replacing a working Sol preset, run a small trial with the same repository snapshot, context, permissions and acceptance test. Count failed attempts and retries, record billed input/output/cache usage, and compare cost per accepted result alongside elapsed time. Keep Sol available until the new route completes text, tools and streaming correctly. Astra is worth switching to when its accepted results or time savings justify the measured extra cost on your tasks; a published leaderboard alone does not establish that.

The worked pricing examples include a first cache write and two follow-ups. The client setup guide covers provider-specific model IDs, Codex Responses settings and migration checks. The independent-index section uses the September 9 report; launch measurements retain their original dates.

TL;DR

• Every rate went up. Input 2x, output 1.67x, cached input 2x. No row is cheaper.
• Independent results now support broader gains. Test your own task mix before migrating.
• The tier naming is gone. No Sol, Terra or Luna on GPT-6. It is Astra and Astra Pro.
• Context is roughly unchanged, 1.05M to 1.05M, same 128K output ceiling.
• GPT-5.6 Sol is still available and still half the input price.

The Price Comparison

Both figures are the live Ofox catalog on 5 September 2026, where Astra’s rates match OpenAI’s list. The full Astra rate card has the Fast-mode and per-task numbers behind these rows.

An input rate that doubles is a bigger deal than it looks for agent workloads, because agents are input-heavy: every turn resends the system prompt, the tool definitions and the accumulated transcript. Cached input doubling from $0.50 to $1.00 compounds the same way. On a long conversation, measure the input, output and cache categories together rather than assuming one column dominates every bill.

What the Extra Money Buys

OpenAI’s launch benchmarks cluster in one place, and it is not general chat:

Computer use, terminal automation, site reliability, exploit discovery. That is a coherent story about what this model is for: these figures motivate task-specific testing, and the workloads it was tuned for are the ones that run long, call tools repeatedly and either succeed or fail at a task rather than producing text a human grades.

Two of those numbers carry qualifiers worth keeping attached: ARC-AGI-3 at 99.9% is with an adapter harness, and Humanity’s Last Exam at 57.2% is with tools.

The adapter qualifier is doing heavy lifting. On ARC Prize’s own standard harness the same model scores 62.7%, and the Sol figure printed beside it on the launch scorecard was measured on a different harness again — OpenAI’s own earlier estimate put Sol near 30% under adapter settings. Our review has the full grid and what it means for reading generational claims.

Independent results updated September 9

Artificial Analysis’s September 9 report, checked September 16, supersedes the older figures previously quoted here. At max effort, Astra scores 53 on its Intelligence Index, six points above Sol and tied with Fable 5.1. In its Coding Agent Index, Astra scores 62, seven points above Sol and tied with Fable 5.1. The reported Astra costs are $3.26 per task on the former, about 60% above Sol, and $7.09 on the latter, about 15% above Sol.

Those are benchmark-specific costs, not an API tariff or a forecast for your repository. The report also records a DeepSWE regression versus Sol. Test the work you actually need: an aggregate gain does not mean every task improves. Do not compare the old and new index numbers as if the measurement scale were unchanged.

The Naming Changed, and Configs Will Notice

GPT-5.6 shipped as Sol, Terra and Luna. GPT-6 ships as Astra and Astra Pro.

That matters operationally more than aesthetically. Anything that builds a model string by pattern, a router that appends a tier suffix, a config template with gpt-5.6-{tier}, a fallback chain that assumes three tiers exist, will not find a GPT-6 equivalent. The API model ID is simply gpt-6-astra.

This is the same class of break that GPT-5.6’s bare model ID caused on gateways, where gpt-5.6 without a tier returned 404 on routes that expected an explicit tier. The lesson repeats: model naming schemes are not stable across generations, and code that generates model strings rather than reading them from a catalog breaks on every launch.

When to migrate, and when to keep Sol

Trial Astra if Sol falls short of a concrete acceptance criterion, whether the task involves reasoning, coding or tools. Switch only when the accepted results or time saved justify measured cost. Keep Sol if it already passes reliably at a lower total cost. Do not assume a higher input rate predicts a whole agent bill: output, caching, retries and tool use also matter.

Changing a model string is not a complete migration. Check the route, supported parameters and tool continuation with the setup guide, then compare a bounded task using the same repository snapshot. Compare Fable where it fits your workload, but do not use the older claim that it leads Astra by five index points as a current selection rule.

Sources

• https://artificialanalysis.ai/models/gpt-6-astra
• https://artificialanalysis.ai/models/gpt-6-astra-high
• https://llm-stats.com/models/gpt-6-astra
• https://ofox.run/models/openai/gpt-6-astra
• https://ofox.run/models/openai/gpt-5.6-sol

GPT-6 Astra benchmarks are OpenAI’s launch figures from 3 September 2026. Rates and context lengths for both models were read from the live Ofox /v1/models endpoint on 5 September 2026, the day GPT-6 Astra was listed. Independent index scores use the September 9 report, checked September 16.