Grok 4.5 arrives in GitHub Copilot, extending a model-agnostic strategy across all four frontier labs

Four Labs, One Picker: GitHub Copilot Completes Its Frontier Model Lineup

GitHub Copilot now hosts AI models from all four major frontier labs after xAI's Grok 4.5 joined the lineup on July 28. The addition finished a seven-day sequence that also brought in Anthropic's Claude Opus 5 on July 24 and Google's Gemini 3.6 Flash on July 21, alongside the OpenAI models already available. 1

The individual models matter less than the pattern. GitHub is turning Copilot's model picker into the real product, and the choice of which AI handles your code is becoming a per-task decision rather than a platform commitment. A developer who wants different models for different jobs can switch between them inside the same editor, paying each lab's listed rate for whatever they use.

Gemini 3.6 Flash: Google's Token-Efficiency Play

Google's contribution arrived first, on July 21. Gemini 3.6 Flash targets web and app developers who need fast iteration on extended agentic workflows. It lets developers tune reasoning depth and fire off multiple tools simultaneously. In GitHub's early testing, the model outperformed its predecessor Gemini 3.5 Flash on both completion rates and token economy. 2 It is available to Copilot Pro through Enterprise plans.

Claude Opus 5: Anthropic's Deep-Reasoning Entrant

Anthropic followed on July 24 with Claude Opus 5, aimed at lengthy coding sessions that demand careful sequential reasoning and multi-tool coordination. GitHub's testing found the model adept at self-directed code modifications and verifying its own output, with less wasted computation on complex tasks. 3 Claude Opus 5 skips the entry-level Pro tier: it starts at Copilot Pro+ and extends through Business and Enterprise. The model ships with guardrails against high-risk cyber content that can interfere with legitimate security work, which is itself an argument for keeping multiple models on standby.

Grok 4.5: xAI's Context-Window Contender

Grok 4.5, added July 28, brings a context window of up to 500,000 tokens and the ability to process both text and images, with three reasoning-intensity levels to choose from. 1 For developers, the context size means large codebases can fit into a single prompt, and multimodal input opens up workflows that pair code with screenshots or design files. GitHub's testing highlighted Grok 4.5's strength in terminal-driven coding scenarios where tools run in parallel and speed matters. Like Gemini 3.6 Flash, it is available from the Copilot Pro tier upward.

Independent assessment outside GitHub tracks with this positioning. Grok 4.5 is described as a coding-first model built for real repository work, agentic tool use, and long-context reasoning, priced at $2 per million input tokens and $6 per million output tokens. 4

The Play: Distribution Over Models

Every model added this week carries the same billing structure: usage-based pricing at each provider's standard list rate. 1 3 2 For Business and Enterprise plans, administrators must enable each new model individually before teams gain access, with policies off by default.

The economics point to a clear bet. If frontier models are converging toward commodity pricing, the durable advantage belongs to whoever controls the developer's daily workflow. GitHub holds the repository, the editor integration, the command-line interface, and the cloud agent. Models become interchangeable inputs feeding those surfaces.

For investors, that reframes the AI tooling thesis. The question shifts from which lab wins any given model generation to which platform reduces switching friction to near zero. GitHub's answer is to stock every shelf and let developers reach for whatever fits the job.

For developers, the constraint is no longer access to frontier models. It is which model matches which problem. Grok 4.5 when a full codebase and a screenshot need to go into one prompt. Claude Opus 5 for a multi-hour refactoring session that requires methodical self-checking. Gemini 3.6 Flash when token costs and iteration speed dominate. OpenAI's models for general-purpose work in between.

The single-vendor AI coding assistant is starting to look like a subset of what Copilot offers, not a competing product. GitHub's wager is straightforward: the platform that refuses to pick a model becomes the one developers cannot leave.