IBM Release Notes

Watch the release video for the latest updates in IBM® watsonx Orchestrate®.

Region availability

The following table shows the rollout status of this release across all supported regions. The deployment date indicates when the release became available in a given region. Regions with an em dash (—) have not received the release.

At a glance

This release delivers Observability for AWS GovCloud, custom plugin controls, LLM-as-a-Judge sampling configuration, workspace filtering and personalized layouts in the control plane dashboard, MCP server registration and collection sharing in the catalog.

MCSP 2.0 migration in Northern Virginia (us-east-1)

As part of the MCSP 2.0 migration, the authentication experience and API key management workflow change. Update your API keys after migration to maintain access to your integrations and applications.

The following changes apply to your environment:

• Enter your email address to sign in. Based on your environment configuration, watsonx Orchestrate directs you to your organization's identity provider (IdP).
• User management moves from the watsonx Orchestrate user interface to the IBM SaaS Console. Manage users from the IBM SaaS Console after migration.
• Generate and manage API keys from the SaaS Console. Existing API keys no longer work after migration. Generate a new API key from the SaaS Console to restore access.

Observability is now available on AWS GovCloud

Availability:

This feature is now available on AWS GovCloud deployments.

The Observability experience is now available on AWS GovCloud tenants. The legacy Analytics experience is no longer supported on AWS GovCloud.

Apply custom controls to agents using Python, OpenAPI, and MCP plugins

You can now create custom controls that enforce organization-specific policies on agent conversations using plugin logic that you supply. Use custom controls when no built-in control covers your requirement, for example, a domain-specific data redaction rule or a proprietary compliance filter.

You can:

• Back a custom control with a Python toolkit, an OpenAPI tool, or an MCP tool already imported into your tenant
• Run the control before the agent processes a user message (agent_pre_invoke) or after the agent generates a response (agent_post_invoke)
• Assign the control to one or more agents from Manage > Controls

Note:

Custom controls can only be assigned to agents. Python toolkit–backed controls are available to Premium tenants only.

Configure sampling rates for LLM-as-a-Judge evaluators

You can now control LLM-as-a-Judge (LLMaaJ) evaluator activation and configure trace sampling percentages for production monitoring.

You can:

• Toggle individual LLMaaJ evaluators on or off from the Evaluators settings panel under Controls > Enterprise Controls > Analytics
• Set a tenant-level default sampling rate between 3% and 20% that applies to all active evaluators
• Configure per-evaluator sampling rate overrides for metrics that require more or less frequent monitoring

Filter control plane dashboard data by workspace

Availability:

This feature is available on IBM Cloud deployments that have private workspaces configured.

You can now filter control plane dashboard metrics by workspace using the workspace selector in the toolbar. Select any private workspace, the Global Workspace, or All workspaces — metrics across all tabs reload to reflect your selection.

Personalize your control plane dashboard layout

You can now customize the control plane dashboard from Customize view in the toolbar. Your layout is stored per user and does not affect what your teammates see.

You can:

• Reorder dashboard tabs and add, reposition, resize, or remove widgets using the Widget library sidebar
• Add section headings to group related widgets into labeled rows
• Restore the original layout at any time using Restore all dashboards to defaults

Register remote MCP servers in the catalog

Availability:

This feature is available to admins only. Builders can discover and use registered servers but cannot register them.

Admins can now register remote Model Context Protocol (MCP) servers directly in the catalog, making their tools available to all builders in the tenant alongside IBM-provided servers.

You can:

• Register a remote MCP server from Catalog > Add to catalog using a three-step panel: provide server details, configure a connection for authentication, and review available tools before adding
• Tag registered servers as By: My organization so builders can filter catalog results to see only your organization's assets
• Support Streamable HTTP and Server-Sent Events (SSE) transport types
• Version each registration, updates create a new version without overwriting previous ones
• Edit or delete a registered server from its catalog detail page

Review evaluation metrics for MCP servers in the catalog

You can now open the Evaluation details tab on any evaluated MCP server in the catalog to review quality metrics before you add the asset to your solution. Evaluated assets display a Certified badge on their catalog tile.

MCP server metrics cover:

• Conformance — Protocol adherence
• Integrity — Tool description coverage
• Safety — Tool description safety and input validation pass rate
• Security — Default credential rejection rate

Note:

An admin runs evaluation via API for assets that require certification. The Evaluation details tab appears only when evaluation results are available for the asset.

Publish skills to the catalog

Administrators can now publish skills to the catalog by uploading a SKILL.md file. Skills contain instructions that guide an agent's behavior for specific tasks, domains, or workflows. Each update creates a new version, and previous versions are preserved.

Availability:

Publishing skills is available to admins only. Builders can discover and use published skills.

Preview the new catalog experience

The updated catalog UI is now available as a preview in all data centers. You can view the new catalog experience, which aligns with the control plane UI design, from the Switch To Catalog Next button in the catalog.

Share catalog collections across tenants

Administrators and builders can now organize agents, tools, and MCP servers into collections, publish them, and share them with other tenants. When a recipient tenant accepts the sharing request, the published assets become available in that tenant.

You can:

• Create a collection and add agents, tools, and MCP servers
• Publish a collection when the assets are ready to share
• Share a published collection with one or more tenants by entering the target tenant ID
• Track the status of sharing requests from the collections management page

Select the ElevenLabs v3 Conversational model for voice agents

You can now select the ElevenLabs Eleven v3 Conversational text-to-speech model in your voice configuration. Voice configurations offer two ElevenLabs models with different performance characteristics.

You can:

• Select Eleven Flash v2.5 (default) for real-time voice interactions, supports 32 languages with approximately 75 ms latency
• Select Eleven v3 Conversational for high emotional range and contextual understanding across 70+ languages
• Build multilingual voice agents by verifying that your LLM, speech-to-text model, and ElevenLabs voice all support your target languages

Extended form-based input types and validation for multi-turn user interactions

Six new input types are now available in Collect from user nodes — Date, Time, Date/Time range, List, User, and Multi-choice, giving you richer structured input options for multi-turn interactions in flows.

Two new configuration fields are also added: Node name for an internal node identifier, and Agent message for the prompt displayed to the user in chat, with support for flow variable insertion.

Insert flow variables into user activity messages and form labels

You can now reference flow variables directly in the message text of Present to user nodes and in form field labels within Collect from user nodes, making it easier to display dynamic, context-aware content during multi-turn interactions.

You can:

• Insert a flow variable into the display text of a Present to user node using the variable picker
• Insert flow variables into the form field Label in Collect from user and Present to user nodes

Copy, cut, paste, and duplicate nodes in Flow builder

You can now copy, cut, paste, and duplicate nodes in Flow builder to reuse configurations and reorganize workflow logic faster. You can also cut, copy and paste nodes between two separate agentic workflows.

You can:

• Copy nodes from one Flow builder to another
• Paste a node directly into a connector line to insert it into the sequence, or paste it onto the canvas
• Cut a node and have the surrounding connector lines reconnect automatically
• Duplicate a node to create a copy without using the clipboard

Note:

Copy, cut, paste, and duplicate are supported for all node types except Start and End nodes.

Schedule agents and flows at any frequency down to every 5 minutes

Execution rate limits for scheduled agents and agentic workflows have been removed. You can now schedule runs as frequently as every 5 minutes using the chat interface or directly via MCP tools.

View files directly in chat

You can now view files returned by agents directly in chat without downloading them, using the inline file viewer in the chat interface.

You can preview pdf, images, and .csv files inline in chat.

Explore intelligent document processing with a preconfigured agent

Every new IBM watsonx Orchestrate tenant now includes a preconfigured Document Processing agent for testing document classification, extraction, and Human-in-the-Loop (HITL) review without creating any workflows. You can also inspect the reference workflow in read-only mode from Active inventory > Agentic workflows.

Extract document content as plain text, markdown, or HTML from the text extractor node

The text extractor node now supports three output formats — plain text, Markdown, and HTML. Select one or more formats in the Extracted content types section of the node configuration panel to match your downstream processing needs.

Upcoming change: Tenant asset limits

Starting with the end-of-October release, each tenant will have default limits on the number of resources they can hold: 1,000 tools (including Python, OpenAPI, remote MCP, and local MCP tools), 1,000 native agents, 5 live + 1,000 draft Python toolkits, and 10 live + 10 draft LangGraph agents.

Existing resources are not affected. Tenants that already exceed a limit will not have resources deleted, but will be unable to add new resources until their count falls below the limit. To free up quota, delete unused assets from your workspace, or use the ADK CLI or platform API. If your use case requires a higher limit, contact the product team.

API updates

Interact directly with external A2A and Chat Completions agents via API

You can now register an A2A-compatible or Chat Completions external agent on the watsonx Orchestrate platform and interact with it directly using the REST API, without routing it through a collaborator.

Other enhancements

The following table lists enhancements available in this release:

• Enhanced performance and analytical accuracy in the Control Plane agent: The conversational agent in the control plane dashboard is upgraded with faster response times, greater accuracy, and enhanced analytical capabilities across your agent fleet and governance metrics. Key improvements include: streamlined instructions and skill configurations that reduce prompt token overhead and improve response latency; improved data resolution logic for accurate active agent and usage metrics; improved statistical filtering that removes low-volume outliers when diagnosing error spikes and failure rates; and better disambiguation between agents that lack evaluation runs and active agents that are missing test cases.
• Enable multi-turn interactions in phone and Genesys Bot Connector: Phone and Genesys Bot Connector now support multi-turn interaction schema v2, giving you the same multi-turn interaction capabilities in those channels as in the chat UI.
• Run up to 20 paths or items concurrently in Parallel and For each controls: Parallel controls and parallel For each controls now have documented concurrency limits. A Parallel control runs up to 20 paths concurrently; additional paths are queued until active paths complete. A For each control in parallel execution mode processes up to 20 items concurrently, with a total list limit of 500 items. These limits are fixed and cannot be adjusted. Learn more: Add a For each control and Add a parallel control
• Run agentic workflows with reduced latency: Agentic workflows now start faster and transition between user activities and automated nodes with reduced latency. Improvements apply to all node types, including Python toolkit, API, MCP, generative AI, decision, and document processing nodes.
• Enable or disable Agent Summarization for agentic workflows: Builders can now configure Agent Summarization on an agentic workflow. When enabled, agent instructions control the workflow summary displayed in the channel. When disabled, workflow outputs are displayed in deterministic, business-friendly formats such as bullet point lists, tables, and files. Agent Summarization is disabled by default.
• Pin and unpin chat threads in the sidebar: You can now pin up to five threads per agent to a dedicated Pinned section at the top of the sidebar. A notification appears when you reach the pin limit for an agent.
• Watson Speech is not available on Azure deployments: Watson Speech is no longer available as a speech provider option for voice agents on Azure data centers. Supported providers for Azure will appear in the speech provider dropdown.
• Receive clearer error messages from the API proxy: Error messages returned by the API proxy have been improved to be more descriptive and actionable, making it easier to diagnose and resolve issues.
• Required timestamp parameters for Get traces API: The Get traces (/v1/agentops-v3/traces) API now requires both fromTimestamp and toTimestamp parameters in all requests. Requests that do not include these parameters will fail. Requiring a defined time range helps improve query performance and ensures more consistent trace retrieval results.

Documentation updates

See the updates in the IBM watsonx Orchestrate documentation for this release.

Original source

Region availability

The following table shows the rollout status of this release across all supported regions. The deployment date indicates when the release became available in a given region. Regions with an em dash (—) have not received the release.

At a glance

This release delivers a redesigned navigation menu, multi-turn interaction support across messaging channels, agent identity with IBM Verify and Microsoft Entra, content guardrails with custom model selection, improved agentic workflow performance, catalog improvements for discovery and sharing, and new AI Gateway connections for Gemini and Azure AI Foundry.

MCSP 2.0 migration in IBM internal servers

As part of the MCSP 2.0 migration in IBM internal servers, your authentication experience and API key management change. Update your API keys after migration to maintain access to your integrations and applications.

Navigate IBM watsonx Orchestrate with a redesigned navigation menu

IBM watsonx Orchestrate now includes a redesigned navigation menu that organizes platform capabilities into clearly labeled sections, making it easier to find and access the features you need.

Configure agent identity with IBM Verify

Availability:

This feature is available as a private preview.

You can now configure agent identity in watsonx Orchestrate with IBM Verify. Tool calls run on behalf of the authenticated user using a delegated access token that represents both the user and the agent identity, improving traceability and reducing the access scope per execution.

Configure agent identity with Microsoft Entra

Availability:

This feature is available as a private preview.

You can now configure agent identity in watsonx Orchestrate with Microsoft Entra. Tool calls run on behalf of the authenticated user using a delegated access token that represents both the user and the agent identity, improving traceability and reducing the access scope per execution.

Enforce content safety with custom models

You can now evaluate agent and tool content with any model that is registered in your tenant by using the Content Guardrail with Custom Model control. Unlike the standard Content Guardrail control, this control lets you select the model that best fits your organization's safety requirements.

Model control enforcement in traces

You can now inspect model control enforcement decisions directly in the agent trace. When a model control executes during an agent run, watsonx Orchestrate adds a new span to the existing trace showing which strategy ran, how it behaved, and whether it succeeded.

Configure workflow settings from a redesigned settings panel in Flow builder

Flow builder now includes a redesigned Flow settings panel that brings all workflow configuration options into one place, replacing the previous settings side panel.

Update a flow to the latest language model in Flow builder

When a workflow uses a legacy language model and a newer model is available, a notification appears in the canvas header. You can now update the flow engine model directly from Flow builder settings.

Data retention policy for agentic workflows

Completed and failed agentic workflow execution runs are now automatically deleted after 30 days. This policy applies to both Draft and Live agentic workflow versions, including runs visible in Flow Inspector and runs that scheduled workflows create.

Reduce voice latency with the Voice optimized reasoning engine

A new Voice optimized option is now available in the Reasoning engine settings of your voice configuration, allowing you to select a lower-latency inference API for voice interactions.

Track agent reasoning steps in sequence during execution

The chat UI now displays reasoning steps sequentially as they occur, giving you a clearer view of what the agent is doing throughout execution.

View conversation history from evaluation results

Availability:

This feature is available only for AWS deployments. It is not currently supported in the AWS GovCloud environment.

Review the complete conversation flow for each test case directly from your evaluation results. Access the full chat history to understand why a test succeeded or failed, and investigate issues by seeing exactly what happened and which tools were called.

Add catalog tools directly to an agent

You can now add a Python tool to an agent directly from the Catalog, so you can discover and configure tools without leaving the catalog.

Detect signatures in documents with intelligent document processing

Availability:

This feature is available as a public preview.

You can now enable signature detection for intelligent document processing (IDP), automating workflows that involve documents requiring a signature, such as loan, mortgage, and lending processing.

Connect AI Gateway to Gemini Enterprise Agent Platform

Availability:

This feature is available as a preview. It is not available in the following regions:

• AWS: Northern Virginia, Northern Virginia Regulated, AWS GovCloud
• IBM Cloud: Paris, São Paulo, Dallas, Washington DC

You can now connect AI Gateway to your Google Cloud project to discover and import agents registered in Gemini Enterprise Agent Platform.

Connect AI Gateway to Microsoft Azure AI Foundry

Availability:

This feature is available as a preview. It is not available in the following regions:

• AWS: Northern Virginia, Northern Virginia Regulated, AWS GovCloud
• IBM Cloud: Paris, São Paulo, Dallas, Washington DC

You can now connect AI Gateway to your Microsoft Azure AI Foundry workspace to discover and import agents registered in your Azure environment.

Other enhancements

The following table lists enhancements available in this release:

• Get notified when IBM or partners publish new assets to the catalog: You can now receive notifications when IBM or partners publish new assets to the catalog, making it easier to discover new capabilities as they become available.
• Improved agentic workflow performance: Agentic workflows benefit from reduced latency, and improved throughput for Script Node (or) Logic Block Node under concurrent load.
• Search chat history by thread title: You can now search for a previous conversation by its thread title from the chat history panel, making it faster to return to a specific conversation.
• Assess catalog agents faster with marketing descriptions on tiles: Agent tiles in the catalog now display a concise marketing description, making it faster to assess an agent's purpose without opening the detail page. Hovering over the tile description shows the full marketing description in a tooltip. For agents that do not have a marketing description set, the tile continues to display the existing functional description.
• Enable multi-turn interactions in Microsoft Teams, Slack, and WhatsApp: Microsoft Teams, Slack, and WhatsApp channels now support multi-turn interaction schema v2, giving you the same multi-turn interaction capabilities in those channels as in the chat UI.
• Control Plane chat experience visual enhancements: The Control Plane chat experience has been redesigned to provide a cleaner, more intuitive, and visually consistent way to interact with the Control Plane agent. Updates improve chat usability, message readability, visual feedback, and overall experience across supported layouts and themes.
• Control Plane dashboard functional improvements: The Control Plane dashboards introduces a more consistent, accessible, and performant dashboard experience. Dashboard visualizations and widgets are now standardized across tabs, empty states provide clearer guidance when data is unavailable, and dashboard loading is optimized to improve responsiveness. You can also switch between chart and table views on selected visualizations for greater flexibility when analyzing data.

Documentation updates

See the updates in the IBM watsonx Orchestrate documentation for this release.

Original source

The first Nighthawk r2 QPU is now available

Premium, Flex, and Pay-As-You-Go Plans now have access to ibm_phoenix, the first IBM Quantum Nighthawk r2 QPU and our fastest processor yet. ibm_phoenix is a high-connectivity, square-lattice 120-programmable-qubit device that delivers up to 25× higher throughput than Heron while matching its fidelity.

The gains come from a newly introduced independent high-speed reset that returns every programmable qubit to the ground state faster and more completely. As a result, the default repetition delay is 1 µs (previously 250 µs), which pushes throughput above 100,000 Max Circuits Per Second (MCPS). For more about how IBM tracks quantum hardware performance across metrics of scale, quality, and speed, see this blog.

The increased speed and connectivity make it possible to do more usable work from the same capacity, with the largest, most demanding workloads benefitting the most. Larger experiments, broader parameter sweeps, and methods that were once too costly at scale are now practical to run routinely.

ibm_phoenix is also built for the next era of quantum error correction research. Alongside high-speed reset, it supports mid-circuit reset and dynamic circuits for real-time control over error correction.

Original source

Updates to IBM Quantum Platform

A recent update to IBM Quantum Platform makes it easier to understand your job results, unlocks powerful new circuit design capabilities in Composer, and keeps you informed about your usage limits — so you can focus more on your quantum work.

Job results & workloads

• Bit-by-bit outcomes view — a new way to view Sampler measurement results, breaking down outcomes bit by bit instead of by full state. Useful as an alternative to the Measurement outcomes view when results contain many distinct states.
• Circuit timing chart — shows a scheduled timeline of a circuit's operations (gates, channels, durations). Available for Sampler and Executor jobs. Requires enabling timing data retrieval.
• Chunk/span execution diagnostics panel — a new panel on job details showing chunk- and span-level execution diagnostics. See Sampler input/output for what this data means and how to enable it when submitting jobs.
• Device calibration data at a glance — job details now show the device calibration data from the time of submission or execution via a "View calibration history" link.
• Executor results visualization — job details pages now render results for Executor jobs. See the Directed execution model (beta) guide for more on Executor.
• Dynamic circuits now render on job details — with OpenQASM 3.0 support in Composer (see more below), circuits that previously failed to render (for example, those using dynamic circuits) now display correctly on the job details page.
• Histogram improvements — faster loading, smoother zoom in/out, and a new selection tool for narrowing in on specific result ranges.
• qiskit-ibm-runtime code snippets in job results — copy-ready code to retrieve a job by ID; for Executor jobs, an additional snippet lets you replicate the job.
• Custom date range filter added to the workloads table.
• Clickable instance names and tags in workload detail pages, linking to the related resource.

Navigate to your results on the Workloads page to see these improvements.

Composer: OpenQASM v3.0 support

Composer now supports OpenQASM v3.0, adding:

• Gate modifiers — ctrl, inv, pow, etc.
• Loops — for and while
• Conditional logic — if/else
• Box annotations — group and label sections of a circuit

Also included: smoother rendering for large circuits, and a fix for PNG exports failing on large circuits.

Visit the Composer to explore this new update.

Instance usage limits

• Premium and Open Plan instances that hit their usage limit now display the exact date access resumes, based on the rolling usage window.

See more on your Instances page.

Original source

Qiskit Runtime rename

• Qiskit Runtime service is now IBM Quantum Compute Service — This is a naming change only and does not require any code changes or updates to existing applications. You might notice the new name throughout the user interface and documentation.

Historically, the term Qiskit Runtime was used to refer to three related but distinct concepts.

Any hardware provider that wants Qiskit users to run workloads on its quantum computers must provide two provider-specific components:

• A Qiskit SDK extension, which enables users to submit jobs to the provider. In IBM's case, this is the open-source qiskit-ibm-runtime library. Rather than communicating directly with the quantum computer, it submits jobs to a cloud API.
• A runtime execution environment, which runs on the quantum computer itself. This component understands Qiskit primitives, translates them into hardware-specific operations, and executes them on the quantum hardware. In IBM's case, this is IBM Qiskit Runtime.

Between these two components sits a third element: the managed cloud service that receives jobs from the qiskit-ibm-runtime SDK and orchestrates their execution. This service is responsible for capabilities such as authentication, scheduling, queue management, workload routing, usage tracking, limits, and analytics. It does not process Qiskit primitives itself; instead, it dispatches workloads to the appropriate quantum computer, where the runtime execution environment performs the actual execution.

Historically, this cloud service was also called Qiskit Runtime Service, even though it is conceptually different from the runtime itself. This dual use of the term Qiskit Runtime — to describe both the execution environment and the managed cloud service — blurred the architecture and made it difficult to distinguish the open-source components from the managed cloud offering.

Renaming Qiskit Runtime Service to IBM Quantum Compute Service resolves this ambiguity. The resulting architecture becomes much clearer:

• qiskit-ibm-runtime: the open-source Qiskit extension that submits workloads.
• IBM Quantum Compute Service: the managed cloud service that authenticates, schedules, manages, and routes workloads.
• IBM Qiskit Runtime (or IBM Qiskit Runtime Execution Environment): the execution environment running on the quantum computer that processes and executes Qiskit primitives.

Original source

Store instruction

• Support for the Store instruction on IBM® QPUs - Now you can simplify complex classical computations in your quantum circuits with the Store instruction.
• Compute a classical expression mid-circuit and save the result to a variable or register, which can then immediately control subsequent quantum gates. Instead of deeply nested inline expressions, you can break computations into named intermediate steps, so that complex classical feedback logic is easier to read and maintain.
  Common use cases include the following:
• Dynamic circuits with mid-circuit measurements and classical feedback
• Long-range entanglement preparation
• Complex conditional logic based on measurement outcomes
  To use the store instruction, first ensure you're running the latest version of qiskit-ibm-runtime (pip install --upgrade qiskit-ibm-runtime), then see the documentation for more information and an example.
  Note that before submitting jobs, you should verify compatibility with each primitive's options in the documentation: Sampler options | Estimator options | Executor options.

Original source

Classroom accounts

• New Classroom Accounts: Professors can now request a Classroom Account to manage and facilitate access for their students. With Classroom Accounts, students can use Open Plan resources without needing to provide personal credit card information.
• New export options now available in IBM Quantum: Introducing new export capabilities that make it easier for account administrators to analyze, share, and work with their data.

New in the Analytics page:

• You can now export charts in SVG or PNG format.
• You can export all analytics data, including the data behind the charts, as CSV.

Find the export menu at the top-right corner of each card on the page.

New in the Access management page:

• Export both the Users table and the Access Groups table in CSV format.

Find the export menu at the top-right corner of the table.

• New resources and hardware for Open Plan users: The IBM Quantum Open Plan now features new learning resources, new hardware, and a special one-time promotion for eligible users.
• A new course, Designing and leading quantum projects, is now available. This course covers key topics for users looking to develop or secure funding for a quantum project, including early-stage planning, team roles and responsibilities, identifying meaningful use cases, defining measures of success, and a guide to grant writing best practices.
• The Heron r2 ibm_kingston —one of our highest-performing quantum computers—is now available to all Open Plan users.
• Log 20 minutes of compute in any 12-month period and you can opt in to a special one-time offer: 180 minutes for the next 12 months.

With expanded runtime and new educational resources, users can run everything from introductory circuits to advanced hybrid workflows.

These efforts reflect our commitment to open access, open education, and open science in quantum computing. Read more: Doubling down on open-access quantum computing.

Original source

The EU gets its first Heron r3

Maintenance has been completed on ibm_aachen and the QPU is now back online in the eu‑de region.

ibm_aachen has also been upgraded to be the first QPU in the EU to feature the latest Heron r3 processor. It is the third r3 in our global fleet, after ibm_boston and ibm_pittsburgh in the US.

Heron r3 continues to deliver improved error rates and overall system performance compared to Heron r2, demonstrating more reliable execution and better results for workloads. For example, on ibm_aachen, we observe a large boost in 𝑇2 (phase coherence time) from 150 µs to 250 µs. You can view metrics like these for all QPUs in the shared fleet on the Compute resources page.

Original source