You asked the AI Assistant to find the root cause of a cascading failure across six services. It traced the issue to a misconfigured retry policy before you finished your coffee.

That was on GPT-5. Today, it does it better.

Starting March 31, 2026, the AI Assistant built into IAPM runs on GPT-5.4, the latest model family from OpenAI, deployed on Azure AI Foundry. This is not a minor version bump. GPT-5.4 brings stronger multi-step reasoning and adaptive logic, which translates directly into sharper root cause analysis, more accurate code-level diagnosis, and more coherent multi-turn investigations during incidents.

What Changed

GPT-5.4 brings stronger multi-step reasoning and improved adaptive logic compared to the previous GPT-5 generation. For the AI Assistant, that translates directly into better performance on the tasks SREs and platform engineers care about:

• Root cause analysis across complex service dependencies
• Multi-hop trace correlation connecting frontend errors to backend database queries
• Code-level diagnosis with full workspace access to your repository
• Agentic orchestration where the assistant plans, executes, and iterates on multi-step investigations

These are the heavy operations, the ones you reach for at 3am when the pager fires.

Right Model, Right Task

Not every question needs the most powerful model. Asking "show me the slowest queries in the last hour" is a different task than "explain why latency tripled after the last deploy and suggest a fix."

The AI Assistant handles this automatically. Every interaction gets routed to one of three operation tiers based on task complexity. Which model actually runs each tier depends on your plan (see the plan table below); on the Analyze and Fuse plans it looks like this:

In practice, the majority of interactions are simple lookups and tool calls that route to the Light tier, keeping costs low and reserving the heaviest models for the moments that demand them. You never have to think about model selection. The assistant picks the right tool for the job, and the energy dashboard shows you exactly what was used and why.

How It Maps to Your Subscription

Your IAPM plan determines which models fill each slot in the assistant's ladder:

Every plan runs the same skills, and the premium GPT-5.4 reasoning model on hard problems is what the Analyze and Fuse plans add. Start and Visualize run the fast GPT-5.4-mini model across the ladder: a fully working assistant for every task type, including root cause analysis, that simply does not escalate to the premium model. Analyze brings GPT-5.4 to heavy and medium work, and Fuse runs it across the board.

The energy you have controls how much you use, not which skills you can reach. Usage is metered as a token budget per 6-hour window, not a daily query count. See the pricing page for full details on each plan.

What This Means for Your Workflow

If you are already using IAPM, the upgrade is automatic. No configuration changes. No new setup. The next time you ask the AI Assistant a question, it is running on GPT-5.4.

Here is what you will notice:

Sharper root cause analysis. GPT-5.4's improved reasoning means the assistant connects the dots across distributed traces more reliably. When a cascading failure spans multiple services, it identifies the origin with fewer false leads.

More precise code-level fixes. The AI Assistant has full workspace access to your codebase. It can read, search, and suggest modifications. GPT-5.4's stronger code comprehension means its suggestions are more precise, especially for complex refactors and multi-file changes.

Better follow-up conversations. Ask it to dig deeper. Challenge its conclusions. GPT-5.4 handles multi-turn investigations with better context retention, so you can iterate on a diagnosis without starting over.

Built on Open Standards

The AI Assistant's intelligence runs on GPT-5.4, but its observability runs on OpenTelemetry. Every trace, metric, and log it analyzes flows through open, vendor-neutral protocols. Upgrading the AI model does not change your instrumentation, your data pipeline, or your ownership of the telemetry data.

This is a deliberate architectural choice. Your AI assistant should get smarter over time without locking you into proprietary formats. As models improve (and they will), the assistant improves with them. Your data stays yours.

What's Next

GPT-5.4 is the foundation, not the ceiling. The assistant's abstraction layer is provider-agnostic by design. As new models prove their value for observability workloads, we will add them to the ladder. The routing system ensures cost stays predictable regardless of which models power the backend.

In the near term, we are validating GPT-5.4-nano as a future Light-tier cost optimization on simple operations. We are also evaluating multi-provider strategies to give you the best model for each task, regardless of who built it.

The AI Assistant does the work. You own the outcome.

Try It Today

Every IAPM plan includes the full AI Assistant. Not a limited version, not a feature-gated tier. Every plan, from free Start to Fuse, gets every Tessa capability: root cause analysis, code review, multi-step autonomous analysis, the full skill set. What changes across plans is how much you use it and how smart it runs on hard problems, never which skills you can reach. The one exception is Agentic Mode, an integration surface explained below, and even that is an energy boundary rather than a skill you lose.

• Start (Free): the full assistant on our fast everyday model, with a recharging usage budget per 6-hour window
• Visualize ($20/node): the full assistant on a stronger model ladder, with more usage headroom
• Analyze ($45/node): a heavier model on hard problems, unlimited usage
• Fuse ($60/node): the top model throughout, unlimited usage

Across the four plans, the skills are identical and the model gets smarter on hard problems. You are never permanently cut off the way a quota wall blocks you: on Visualize the model steps down to a lighter, faster mode when the usage budget runs low and recovers when the 6-hour window recharges, and on the free Start plan the assistant pauses when the window is spent, then resumes at full quality when the window rolls over. Analyze and Fuse are unlimited and never throttle.

Agentic Mode (CLI and IDE integration, deep-research sessions) needs the E5 energy level. E5 is part of a Tessa Personal subscription, a separate axis from your node plan, so it is not unlocked by moving up the node tiers. This is an integration-surface boundary, not a skill limit: the same analysis Tessa does in Agentic Mode, she also does in Chat Mode, which is available on every plan. See Tessa pricing for Personal subscriptions and E5.

Ready to try it? Start free, no credit card required, and you get the full assistant from the first query.

Start Free. Immersive. AI-guided. Full-stack observability. Enter the World of Your Application®.