GitHub Copilot

GitHub Copilot is now a multi-surface AI coding platform spanning code completion and next-edit suggestions, chat, agent mode, Copilot CLI, the Copilot app, cloud coding agents, code review and GitHub-native development workflows. Its September 2026 model lineup materially expands the high-end agentic ceiling. GPT-6 Astra is generally available to Copilot Pro+, Max, Business and Enterprise users across major Copilot surfaces, including VS Code, Visual Studio, CLI, the coding agent, github.com, GitHub Mobile, JetBrains, Xcode and Eclipse.¹ Claude Fable 5.1 is available to the same higher-tier plans, while Gemini 3.8 Flash also reaches Copilot Pro.²³

GitHub is also testing Project HydraFusion, a Copilot CLI research preview that orchestrates models from multiple providers at runtime. HydraFusion can solve a task with one model, cascade from a cheaper draft to a stronger model when needed, or use a separate critique/revision path. It is available through the experimental Copilot CLI surface across Copilot plans, with usage charged according to the underlying models it invokes. GitHub reports favorable quality/cost results in its own controlled evaluations, but those vendor benchmarks are treated as release evidence rather than independent rating inputs.⁴

The workflow layer is also becoming more consistent across environments. GitHub says its newer Copilot harness is generally available in JetBrains, while the Copilot app and CLI now honor administrator-configured content exclusions.⁵⁶ Copilot therefore combines unusually broad IDE distribution with GitHub-native repository, pull-request, review and agent workflows rather than forcing teams into a separate AI-native editor.

Two September changes materially strengthen Copilot's governance and review surface. Enterprise-managed permissions now let Copilot Business and Enterprise administrators centrally classify supported agent operations as denied, approval-required or allowed without prompting across shell commands, file reads/edits and network domains. GitHub states that managed restrictions cannot be weakened by user/workspace settings, auto-approval or stored approvals; the controls are generally available in the Copilot app, Copilot CLI and VS Code sessions using Agent Host.¹⁵¹⁶

Copilot code review also now produces an approval assessment for every review. In public preview, administrators can separately enable Copilot to submit an approving review that counts toward required pull-request approvals; this is off by default and configurable at enterprise, organization and repository scope. New commits dismiss Copilot's approval like a human approval. This improves automation inside branch-protection workflows, but it should not be described as autonomous merge authority: the approval role is opt-in, remains subject to repository rules, and the feature is still preview.¹⁷

The individual lineup remains Free at $0, Pro at $10/user/month, Pro+ at $39/month and Max at $100/month. Paid plans keep unlimited code completions and next-edit suggestions while chat, agents, code review, CLI and other generative surfaces consume GitHub AI Credits. GitHub currently describes Pro as including $15 of monthly total credits, Pro+ $70 and Max $200, with additional credits purchasable on paid individual plans.⁷

Copilot Business remains $19/user/month with 1,900 monthly AI credits per seat and Enterprise $39/user/month with 3,900; organization credits pool at the billing entity level and excess usage is billed separately.⁸ Self-serve Business and Enterprise signups are now being gradually reopened after the earlier pause, with updated seat-payment and account-vetting behavior but unchanged plan prices.⁹

Model availability is no longer uniform across paid tiers. Gemini 3.8 Flash reaches Pro as well as higher plans, whereas GPT-6 Astra and Claude Fable 5.1 require Pro+, Max, Business or Enterprise. Astra and Fable 5.1 are billed at provider list pricing under usage-based billing, so premium-model choice can materially change credit consumption.¹²³

GitHub Copilot continues to have one of the deepest independent exact-product evidence bases in AI coding. G2 reports about 4.5/5 from 277 reviews, Capterra 4.5/5 from 50 reviews, Gartner Peer Insights roughly 4.4/5 from about 467 ratings and TrustRadius 8.6/10 from 80 ratings.^10111213 Recurring strengths are integration, productivity and ease of adoption; recurring drawbacks include inaccurate suggestions and cost/usage friction. These are product-level ratings across models and workflows and should not be attributed to Astra, Fable, Gemini or any other single foundation model.

The September model expansion adds a more complicated governance story. Claude Fable 5.1 requires Anthropic data retention by default so that safety classifiers can operate; GitHub offers zero-data-retention routing only to certain eligible enterprise customers under a time-bound exemption.² Separately, content exclusions are now generally available in the Copilot app and CLI for Business and Enterprise, helping keep excluded files out of those agentic contexts.⁶ GitHub's documentation still notes that agent mode in Copilot Chat inside IDEs does not support content exclusion, so administrators should not assume one exclusion policy protects every Copilot surface.¹⁴

GitHub's claims that Astra, Fable 5.1 or Gemini 3.8 Flash perform strongly in internal testing are useful release context but are not independent benchmark evidence and are not fed directly into the AiToolMap rating equation.¹²³

Copilot is best for individual developers and engineering teams that want AI assistance embedded into existing GitHub and mainstream IDE workflows rather than migrating to a new editor. Pro remains the practical individual entry tier for broad everyday use; Pro+ and Max make more sense when premium models and sustained agent workloads justify higher included usage. Business and Enterprise are strongest where centralized policy, pooled budgets, repository governance and GitHub identity already matter.

It is a weaker fit for users who want simple fixed-cost economics, fully model-agnostic bring-your-own-provider infrastructure, or a single uniform privacy policy across every model and agentic surface. Organizations handling sensitive code should evaluate model-specific retention requirements and the exact surfaces covered by content exclusions before enabling new models by default.