GitHub Copilot opens up to third-party models
GitHub Copilot has always depended on multiple large language models depending on the task at hand. The original public release was built on Codex, an early version of OpenAI's GPT-3 fine-tuned specifically for code generation. Copilot Chat arrived in 2023 running on GPT-3.5, later moving to GPT-4, and since then GitHub has cycled through several base model versions—from GPT 3.5-turbo to GPT 4o and 4o-mini—to balance latency against output quality.
The last year has produced a wave of LLMs with markedly different strengths, and GitHub's position is that the next era of AI-assisted development will be defined not just by multi-model capability, but by multi-model choice. Announced at GitHub Universe, Copilot now lets developers pick between Anthropic's Claude 3.5 Sonnet, Google's Gemini 1.5 Pro, and OpenAI's o1-preview and o1-mini models.
The rollout begins in Copilot Chat. OpenAI's o1-preview and o1-mini are available immediately; Claude 3.5 Sonnet will appear progressively over the following week, with Gemini 1.5 Pro expected in the coming weeks. GitHub plans to extend model selection across Copilot Workspace, multi-file editing, code review, security autofix, and the CLI in the near future. Individual developers can choose per session in VS Code or on GitHub.com, while organization and enterprise admins control which models are enabled for their teams.
Claude 3.5 Sonnet: lifecycle coverage
Anthropic's Claude 3.5 Sonnet is positioned as a strong generalist for software work, performing well across the full development lifecycle—from initial design through bug fixes, maintenance, and optimization. GitHub highlights its facility with complex, multi-step tasks, including legacy codebase updates, refactoring, and new feature work.
Gemini 1.5 Pro: long context and multimodality
Google's Gemini 1.5 Pro stands out for its two-million-token context window and native multimodal support, allowing it to process code, images, audio, video, and text simultaneously. For Copilot, GitHub cites particularly fast response times on routine requests like code suggestions, documentation, and code explanation.
OpenAI o1-preview: reasoning-first models
The o1-preview and o1-mini models represent a shift in OpenAI's lineup toward explicitly reasoning-oriented architectures, a step beyond GPT 4o. In GitHub's own testing with Copilot, these models demonstrated a deeper grasp of code constraints and edge cases, producing more efficient and higher-quality output as a result.
Model choice operates under a single login and subscription, so no additional accounts or setups are required when switching between providers.
Model selection previewed in GitHub Spark
GitHub also unveiled Spark, its AI-native tool for building applications purely from natural language prompts. Sparks are described as fully functional micro apps capable of integrating AI features and external data sources without requiring cloud infrastructure management. The workflow relies on a "creativity feedback loop": users begin with a prompt, watch live previews as the app is generated, review alternatives for each request, and keep automatic version snapshots to compare iterations side by side. Interest in Spark can be registered at gh.io/spark-signup.
