GitHub Added GPT-6.2 to Copilot, and Free Users Cannot Select It
The model list in Copilot keeps growing and the segment of it a free account can touch has not moved, because there is no list to touch.
GPT-6.2 Astra and GPT-6.2 Sol reached general availability in GitHub Copilot on 17 November 2026. Both are available to Copilot Pro+, Max, Business and Enterprise users, and both are billed at provider list pricing under usage-based billing. Copilot Free and Copilot Student get no model picker at all: those plans reach models through auto model selection only, so a new model arriving is not something either plan can choose.1
Each Copilot release this month has widened the model catalogue and left that one sentence untouched. It is worth reading slowly, because it means the gap is not a rate limit that more patience will clear.
What launched, and at what price
Both models are aimed at agentic coding and code review. According to GitHub's changelog, its own testing claims Astra caught significantly more real issues in review than GPT-6 Astra while using noticeably fewer tokens, and that Sol caught nearly twice as many real issues as GPT-6.1 Sol. Those are vendor claims from early testing, not independent benchmarks.
The billing is per token, in GitHub AI credits at one cent each, and the two models sit at genuinely different price points:2
| Model | Tier | Input | Cached input | Cache write | Output |
|---|---|---|---|---|---|
| GPT-6.2 Astra | Default, up to 272K | $8.00 | $0.40 | $10.00 | $40.00 |
| GPT-6.2 Astra | Long context, over 272K | $16.00 | $0.80 | $20.00 | $60.00 |
| GPT-6.2 Sol | Default, up to 272K | $2.00 | $0.10 | $2.50 | $10.00 |
| GPT-6.2 Sol | Long context, over 272K | $4.00 | $0.20 | $5.00 | $15.00 |
Prices measured on 27 October 2026. Vendor plan terms, allowances and prices change without notice, so treat every figure on this page as a dated snapshot rather than a permanent fact. Where a number here differs from the vendor's own pricing page, the vendor's page is correct and the figure above is simply older than it.
All figures are dollars per million tokens. Astra at the default tier costs four times what Sol does on both input and output, and crossing 272K of context doubles every line in the table. The same conversation that is affordable at 250K becomes twice as expensive one prompt later, which is the sort of cliff that does not appear in a model picker.
Astra, Sol and the GPT-6 series also carry a cache write cost that earlier OpenAI models in Copilot do not, so the cached input rate is not the whole caching story. Ignoring cache write understates the true cost of any workflow that keeps a long conversation warm.
What a free account actually gets
GitHub's billing documentation is direct about it. Copilot Free and Copilot Student both have an allowance of AI credits and access to models through auto model selection only. Copilot Free includes 2,000 code completions per month, and Student includes unlimited completions.
| Plan | Price/month | Total monthly AI credits | Model choice |
|---|---|---|---|
| Copilot Free | $0 | An allowance, not itemised | Auto selection only |
| Copilot Pro | $10 | 1,500 | Full picker |
| Copilot Pro+ | $39 | 7,000 | Full picker |
| Copilot Max | $100 | 20,000 | Full picker |
Notice that the paid plans publish their allowances as base credits plus a flex allotment, and the free plan's allowance is described without a figure. That is not a complaint about the generosity of 2,000 completions, which is a real amount of free work. It is an observation about what can be planned. A team evaluating Copilot cannot forecast cost on the free tier because the quantity is not published, and cannot select a model because there is no picker.
The part that surprises people
Astra is the flagship and Sol is the cheaper sibling, and GitHub's own pricing shows Sol at $2 input and $10 output against Astra at $8 and $40. Anyone who assumes the newer number is the better deal is wrong in one direction or the other depending on the task, because the price difference is real work priced differently rather than a premium for a badge.
There is also the long-context threshold to consider. Copilot splits both models into a default tier up to 272K tokens and a long-context tier above it, with the rate doubling at the boundary. In our reading of the pricing tables, an agent working through a large repository can drift across that line without anyone deliberately choosing to, and there is no notification when it happens. The caveat worth internalising is that context is a billing control, not just a capability.
Administrators on Business and Enterprise get a policy lever: new models are enabled automatically unless an administrator has turned off the global default or explicitly disabled them. A model appearing in the picker is therefore the default state, and a company that wants Astra off will have to say so.
How this fits the rest of the month
We have now tracked four Copilot model changes in a single week. Six older models retire on 14 December, and the free and student plans were already confined to auto selection then. The free plan's credit position has not changed either.
The verdict is that this is a good release for anyone on Pro+ and above, and a non-event for a free account, which is the pattern across every Copilot launch this month. The limitations are the auto-select-only restriction, the unpublished free allowance and the doubling at the long-context threshold. Best for Pro+ and Max users doing agentic work who want a cheaper strong model in Sol; not available at all on Free, where the model list is something you read about rather than choose from.
Frequently asked questions
Can Copilot Free users select GPT-6.2 Astra or Sol?
No. Copilot Free and Student reach models through auto model selection only.
Which plans can use GPT-6.2 models?
Copilot Pro+, Max, Business and Enterprise.
How much does GPT-6.2 Astra cost?
$8 per million input and $40 per million output up to 272K context, doubling above that.
How much does GPT-6.2 Sol cost?
$2 per million input and $10 per million output up to 272K, doubling above that.
Is there a cache write cost?
Yes, for GPT-6.2 and the GPT-6 series, on top of cached input.
Can an administrator block these models?
Yes, through the model policy in Copilot settings. New models are enabled by default unless the global default is off.
The availability list, the agentic coding and code review claims and the administrator policy behaviour are from GitHub's changelog entry of 17 November 2026. The per-million-token rates, the 272K tier threshold, the cache write cost, the AI credit allowances and the auto-selection-only statement for Free and Student are from GitHub's Copilot models and pricing documentation and the individual billing documentation, read on 18 November 2026.
Sources
- GitHub Changelog: GPT-6.2 Astra and Sol: github.blog/changelog/2026-11-17-openais-gpt-6-2-astra-and-gpt-6-2-sol-now-available. Used for: GPT-6.2 Astra $8 in / $40 out per million, Sol $2 / $10; doubles above 272K context; cache write cost applies.
- GitHub Docs: Copilot models and pricing: docs.github.com/en/copilot/reference/copilot-billing/models-and-pricing. Used for: Copilot Free and Student get auto model selection only; Free includes 2000 completions per month.
- GitHub Docs: individual billing: docs.github.com/en/copilot/concepts/billing-and-usage/individuals/billing.
All sources are the vendor's own plan or documentation pages, checked on 27 October 2026. Plan terms change without notice, so confirm current limits before you rely on them.