PuppyIP Resource Center
AI Coding Tools 7 min read Published 2026-11-18

GPT-6.2 Astra and Sol in Copilot: Plans, Cache Writes and Costs

GitHub has made GPT-6.2 Astra and Sol generally available to Copilot Pro+, Max, Business and Enterprise, with access rolling out gradually. Check your plan and model policy first, then review token charges: cache writes have a separate rate, and input above 272K has a long-context tier.

GitHub Copilot GPT-6.2 Astra GPT-6.2 Sol AI Credits Cache writes Long context

Service eligibility and regional restrictions

PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.

Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.

Key Takeaways

  • The November 17 announcement lists Pro+, Max, Business and Enterprise. The current support matrix explicitly excludes Pro; eligibility for another model does not establish access.
  • The announcement covers VS Code, Visual Studio, CLI, the coding agent, Copilot app, web, mobile and other IDEs. Rollout is gradual, and all five published IDE minimum-version entries remain TBD.
  • Prices per million tokens separate input, cached input, cache writes and output. Default Astra rates are USD 8/0.40/10/40; Sol rates are USD 2/0.10/2.50/10.
  • Both models support one-million-token context and configurable reasoning. The million-token option is limited to VS Code and CLI, with a separate price tier above 272K input tokens.

What has arrived in Copilot?

GitHub announced general availability of GPT-6.2 Astra and GPT-6.2 Sol in Copilot on November 17, 2026; the support table also marks both as GA. GitHub describes them for agentic coding and code review, positioning Astra for agentic coding and terminal work and Sol for complex agentic coding and in-depth review. Here, an agent is an assistant that can carry out multiple coding steps toward a task.

The announcement includes early test results, but this guide has not reproduced them or established faster, more accurate or more token-efficient results for any repository. This is a GitHub integration announcement. Check Copilot eligibility, clients and billing before use.

Which plans and clients qualify?

GitHub lists Copilot Pro+, Max, Business and Enterprise. The current plan matrix marks Pro as Not included for both models and does not establish Free or Student eligibility. Confirm the actual Copilot plan rather than carrying over the plan list for MAI or another model.

Look in the model pickers listed in the announcement: Visual Studio Code, Visual Studio, Copilot CLI, GitHub Copilot coding agent, GitHub Copilot app, github.com, GitHub Mobile on iOS and Android, JetBrains, Xcode and Eclipse. GitHub explicitly describes a gradual rollout and says to check back if the models are missing.

The current minimum-version table shows TBD, meaning not yet specified, for both models across VS Code, Visual Studio, JetBrains, Xcode and Eclipse. Do not substitute a version requirement from an older model. The announcement gives no completion time for access across all accounts.

Which model policy should a team check?

Copilot Business and Enterprise administrators can manage GPT-6.2 access through the model policy in Copilot settings. Under the announcement’s default model-enablement rule, new models are enabled automatically unless an administrator has turned off the global default or explicitly disabled these models. Have an authorized administrator check the applicable enterprise, organization and model settings.

Permission to use a model and permission to incur extra charges are separate decisions. Members should still confirm the model selected for the current session, organizational data requirements and budgets. Default enablement does not establish that every existing session switched models or that usage is unlimited.

Read all four default rates, including cache writes

At the November 18, 2026 review, GitHub lists USD prices per million tokens, the units used to measure model input and output. The default tier applies at an input threshold of at most 272K, or 272,000 tokens. In the order input, cached input, cache write and output, GPT-6.2 Astra costs USD 8.00, 0.40, 10.00 and 40.00; GPT-6.2 Sol costs USD 2.00, 0.10, 2.50 and 10.00.

Cached input concerns reading reused context; cache writes have a separate listed rate. GitHub explicitly includes both models among OpenAI models with cache-write charges. A low cached-input rate alone is insufficient for an estimate. Do not assume every input hits the cache or produces the same write volume; check actual usage categories separately.

How do million-token context and the 272K threshold differ?

The support matrix lists both models for a one-million-token context window and configurable reasoning levels. Context is the material a session can process. GitHub limits the million-token context option to Visual Studio Code and Copilot CLI; model support does not establish an identical control on mobile or in every IDE.

For the Long context tier above 272K input tokens, the same input, cached-input, cache-write and output order gives Astra rates of USD 16.00, 0.80, 20.00 and 60.00, and Sol rates of USD 4.00, 0.20, 5.00 and 15.00 per million tokens. These are the published threshold and rates. The table reviewed here does not explain whether the threshold calculation applies to excess tokens alone or the whole request, so this guide does not invent a billing algorithm.

GitHub warns that larger context windows or higher reasoning levels consume more tokens and AI credits. Start with regular context and reasoning, increasing them when the task needs it. An available million-token capacity is not a fixed charge for every task.

How do these rates translate into AI credits?

One GitHub AI credit equals USD 0.01. Included personal allowances vary by plan; Business and Enterprise pool per-user allowances at the billing entity level. Additional usage beyond included allowances follows the listed rates and budget settings. Model access does not mean unlimited usage.

Suppose a default-tier interaction uses exactly 100,000 ordinary input tokens and 20,000 output tokens, with no cached input, cache writes or other charges in this example. Astra costs 0.1×8 + 0.02×40 = USD 1.60, or 160 AI credits; Sol costs 0.1×2 + 0.02×10 = USD 0.40, or 40 credits. This is arithmetic under stated assumptions, not a measured bill or a prediction that both models use identical tokens for the same task.

A real task can include multiple model calls and different cache categories. Check included allowances, additional-usage budgets and stop-at-budget settings before starting, then inspect actual usage afterward. Input rates or one fast response alone do not establish the complete task cost.

Check these four things when the models are missing

First confirm the signed-in account and its actual Copilot plan, then check the client and model picker. For a managed account, have an administrator inspect model policies, and retain gradual rollout as a condition. Published IDE minimum versions remain unspecified; a missing option alone does not diagnose a network fault.

If access remains unavailable, record the client version, plan, observation time and non-sensitive error, then use GitHub’s official support process. These sources provide no complete country-by-country availability table or confirmation of a particular account’s current entitlement or remaining allowance. A proxy exit cannot add subscription eligibility, change administrator policies or bypass billing. This guide made no account changes or paid calls.

Sources

Frequently Asked Questions

Does a ChatGPT Plus or Pro subscription give me these models in Copilot?

Check the GitHub Copilot plan separately. This announcement lists Copilot Pro+, Max, Business and Enterprise. OpenAI Work/Codex subscriptions and speed permissions do not establish Copilot entitlements. This integration announcement also does not announce Ultrafast access in Copilot.

Will Copilot code review use the GPT-6.2 model I selected in chat?

That is not established. GitHub’s current billing documentation says Copilot code review selects its model automatically without disclosing it, so per-token costs can vary between reviews. Choosing GPT-6.2 in a chat picker does not identify the model actually used by an automatic code review.

Does this release mean my Copilot allowance has reset?

The GitHub model announcement does not confirm an allowance reset for individuals or organizations. Check the applicable plan and actual usage for access, remaining credits and permission for additional usage. A model launch does not establish a reset for every account, and reset announcements from other platforms do not transfer to Copilot.