JetBrains AI Assistant and Junie Now Support GPT-5

TL;DR

• JetBrains AI Assistant, Junie, and Kineto now support OpenAI's GPT-5 model
• Internal benchmarks show 1.5× to 2× improvements in code quality and task complexity handling over previous OpenAI models
• GPT-5 is now the default model for all three products, available in JetBrains IDE version 2025.2+
• If you're using Junie or AI Assistant in production, this upgrade delivers measurably better code generation without changing your workflow

The Big Picture

JetBrains just flipped the switch on GPT-5 support across its AI product line. AI Assistant, Junie, and the new Kineto platform all default to OpenAI's latest model as of today.

This matters because JetBrains had early access to GPT-5 and ran it through their internal benchmarks before making it the default. They're claiming 1.5× to 2× improvements in code quality and task complexity handling compared to GPT-4. That's not marketing speak — it's the kind of jump that changes what you can realistically ask an AI coding tool to do.

The timing is strategic. OpenAI's iterative deployment approach means JetBrains got to test GPT-5 in real IDE workflows before the wider rollout. According to Olivier Godement, Head of Business Products at OpenAI, this early partnership helped identify where the API delivers the most impact for developer users.

For developers already using Junie or AI Assistant, this is a free upgrade. No new subscription tier, no opt-in process. Update your plugins and you're running GPT-5.

How It Works

GPT-5 integration lives at the model selection layer in JetBrains' AI stack. When you invoke AI Assistant's chat or trigger Junie to handle a coding task, the request now routes to GPT-5 by default. You can still swap models in plugin settings if you prefer a different provider — JetBrains maintains support for multiple LLMs across their tools.

The performance claims are specific to coding tasks. JetBrains tested GPT-5 against scenarios like multi-file refactoring, dependency-aware code injection, and generating production-ready components from natural language prompts. The 1.5× to 2× improvement metric comes from internal benchmarks measuring code quality, correctness, and the model's ability to handle complex, multi-step tasks without hallucinating or breaking existing code.

One example from their testing: Junie was asked to create a hidden snake game as an Easter egg in a footer. GPT-5 didn't just generate the game logic — it identified the correct injection point in a large codebase, accounted for layout constraints and dependencies, and delivered code that worked on the first try. No manual fixes, no "close but not quite" output that requires developer cleanup.

For AI Assistant, the upgrade shows up most clearly in single-prompt complexity. JetBrains demonstrated a prompt asking for an HTML5 page showing what a JetBrains IDE could look like in 2030. GPT-5 generated a fully responsive, styled, interactive prototype from that one sentence. Previous models would have required multiple follow-up prompts or manual adjustments to reach the same result.

The technical advantage comes from GPT-5's improved context handling and reasoning. It can track more variables across a larger codebase, which means fewer mistakes when modifying existing code or integrating new features into complex projects. For spec-driven development workflows, this translates to fewer iterations between the AI's output and what you actually need.

Kineto, JetBrains' new no-code platform, benefits most from GPT-5's frontend capabilities. The model defaults to GPT-5 because it excels at generating single-purpose apps and websites from user prompts. Kineto is still in Early Access, but the GPT-5 integration is live for waitlist participants.

What This Changes For Developers

If you're already using Junie or AI Assistant, the workflow stays the same. The difference is in what you can ask for and how often the first output is usable.

Tasks that previously required multiple rounds of refinement — like "add this feature but make sure it doesn't break the existing auth flow" — now have a higher success rate on the first attempt. That's the practical impact of better context handling. GPT-5 is less likely to ignore constraints you mentioned earlier in the conversation or overlook dependencies in your codebase.

For teams using Junie as a coding agent, the 1.5× to 2× improvement in task complexity handling means you can delegate more ambitious work. Instead of breaking a feature into smaller subtasks to avoid confusing the AI, you can describe the full requirement and trust the output will account for edge cases and integration points.

The cost structure hasn't changed. JetBrains is absorbing the GPT-5 API costs within existing AI Assistant and Junie subscriptions. You're not paying more for access to the new model, which is unusual — most platforms charge a premium for their latest LLM tier.

One caveat: GPT-5 is the default, but you can still switch to other models if you prefer. JetBrains maintains LLM-agnostic architecture across its AI products, so if you have a workflow optimized for Claude or another provider, you're not locked into OpenAI.

Try It Yourself

To access GPT-5 in your JetBrains IDE, you need version 2025.2 or later. Update AI Assistant and Junie plugins via Settings | Plugins. GPT-5 will be selected as the default model in AI Assistant's chat and Junie's agent interface.

If you're testing the upgrade, try a task that previously required multiple iterations. Something like "refactor this class to use dependency injection, but keep the existing API surface unchanged" or "add error handling to this function without changing its signature." Compare the first output to what you'd get from GPT-4 or your previous default model.

For Kineto access, join the waitlist at kineto.dev. The platform is in Early Access, so availability is limited.

The Bottom Line

Use this if you're already on JetBrains AI Assistant or Junie and you're hitting the limits of what GPT-4 can handle in a single prompt. The upgrade is automatic and free, so there's no reason not to try it. Skip it if you've optimized your workflow around a different LLM provider and you're not seeing pain points with your current setup.

The real opportunity here is for teams using Junie in production. The 1.5× to 2× improvement in code quality and task complexity isn't hype — it's the difference between an AI that needs constant supervision and one that can handle multi-step features autonomously. If you've been hesitant to delegate complex tasks to Junie because the output quality was inconsistent, GPT-5 changes that calculation.

The risk is minimal. You can revert to a different model in settings if GPT-5 doesn't meet your expectations. But based on JetBrains' early testing and OpenAI's track record, this is the most capable coding model available in a JetBrains IDE right now.

Source: Junie