GitHub Copilot Cost Accuracy

- AI Explorer: The AI Explorer now defaults to billing-sourced cost data for everyone, giving you AI spend figures that stay consistent with the rest of your billing views.

• AI Explorer: You can now favorite the dimensions and tags you use most in the group-by and filter pickers, pinning them to the top so they are quick to reach.
• AI Signals: Token usage now appears by default in the AI Explorer cost table, shown before Total Cost, so you can see consumption and spend side by side without customizing columns.
• AI Signals: The desktop collector now surfaces branch, pull request, work item, repository, and customer tag fields, so you can group and filter AI spend by the development context that drove it.
• AI Signals: OpenAI Platform and GitHub Copilot spend are now broken out by individual user, with OpenAI costs also attributed to ChatGPT workspace and plan, so you can see who and which team drove each call.
• AI Signals: A grouped account now sees its combined usage even when some member organizations have no billing data, and those members are named so you know which are excluded.
• AWS Connection: Newly linked AWS accounts now reconcile within 15 minutes instead of up to 12 hours, so corrected connection errors and new accounts appear far sooner.
• AWS Connection: You can now test an AWS CUR 2.0 connection before or after setup to confirm CloudZero can read the export, catching misconfigured roles or prefixes before ingestion.
• Cloud Connections: OpenAI cost ingestion now keeps pace with connection size, so large and growing accounts keep ingesting reliably as they scale.
• Cloud Connections: AnyCost billing data can now include custom line-item properties and a cost source, giving you richer detail and attribution in the cost you send to CloudZero.
• Control Plane: You can now rename devices directly from the device list, and access to device management now follows your organization's access controls, so teams see and edit only what their permissions allow.
• Control Plane: The device flyout now resolves parent and collector names, shows host, ownership, and first-seen time, and flags revoked devices, so you can answer connection questions without raw IDs.
• Data Plane: Configuring the data plane Helm chart is simpler, letting you set the upstream and control plane hosts once instead of repeating them across every component.
• macOS Collector: OpenAI, ChatGPT, and Codex traffic is now captured, including credit balances and Responses API calls, so all your OpenAI spend appears alongside your other AI activity.
• macOS Collector: The setup wizard now lets you choose which AI providers to track, so first-run setup only walks you through the steps that apply to your selection.
• MongoDB Connection: Invoice tax now appears as its own line item, so your MongoDB spend in CloudZero matches the total on your invoice.
• Optimize: A new beta recommendation flags Google Cloud Storage multi-region buckets that could save with Rapid Cache, and its titles now lead with the bucket name so you can tell similar recommendations apart.
• Optimize: A new recommendation detects Amazon DocumentDB clusters paying AWS Extended Support surcharges for staying on engine version 3.6, so you can upgrade and stop the extra charge.
• Organization Settings: Org admins can now see their billing status and SSO provider directly in Organization Settings, so they no longer need to contact support to answer basic account questions.
• Strands Agents: AI usage from the AWS Strands Agents framework is now recognized as its own telemetry source, so each model call is counted once and attributed to the provider that served it.
• Telemetry Streams: Standard OpenTelemetry exporters can now send traces to any telemetry source using their default endpoint, so you no longer need source-specific configuration to start streaming.
• Unit Costs: When your organization has no telemetry streams yet, the Unit Cost picker now explains this and links you straight to creating one instead of showing an empty list.

- AI Signals: AI telemetry from Gemini, Amazon Bedrock, and Azure OpenAI hosts no longer creates duplicate connections when your cloud bill is already connected, so AI spend stays under one connection.

• AI Signals: API key alias and other identity fields populate again after a brief gap left them blank; the affected window before the fix is not backfilled.
• AI Signals: Livestream usage that is still awaiting classification now shows as Pending instead of Unknown, so recently ingested activity no longer reads as a failed classification.
• AI Signals: Amazon Bedrock usage sent through the Strands Agents SDK is now correctly attributed to AWS instead of appearing as an unpriced provider at zero dollars.
• AI Signals: AI model costs for context-window and regional model variants now resolve to a clean model name, so their spend appears in your totals instead of quietly dropping out.
• AI Signals: AWS Bedrock spend now groups under a single cloud platform instead of scattering across five names, and a new service breakdown shows which AWS service handled each call.
• Analytics: Search terms and filters no longer carry over when you switch organizations or navigate away, so each dashboard opens with the correct, unfiltered results.
• Anomalies: Anomaly detection now defaults to on for organizations that never explicitly configured it, restoring alerts that were unintentionally silenced.
• Anomalies: Long insight lists now load reliably instead of returning a server error, with results paged so high-volume accounts stay within response limits.
• Anomalies: Switching the granularity on a detail cost chart no longer resets your selected time range; options that do not fit the current range are now disabled instead.
• Anomalies: The same anomaly no longer triggers duplicate Slack alerts, so you receive exactly one notification per anomaly instead of repeated identical messages.
• AWS Connection: A month with incomplete data is no longer reported as complete and is now retried, so AWS cost totals are no longer silently understated when an ingest file fails.
• Azure Connection: Current-month Azure spend is no longer understated by a spurious credit, because invoice reconciliation now runs only after the billing period closes and matches the Azure portal.
• Azure Connection: Azure billing data now keeps updating reliably after a fix to a safety check that could otherwise freeze a billing period on valid data.
• Azure Connection: Temporary failures while retrieving Azure cost reports are now retried automatically, so a healthy connection no longer shows an error status several times a month.
• Cloud Connections: Billing entity and cloud provider on AnyCost connections now save and display correctly when created or updated, instead of failing or silently reverting.
• Cloud Connections: OpenAI cost ingestion now completes for organizations with thousands of projects, instead of timing out and restarting from scratch without ever producing data.
• Cloud Connections: The billing entity on a connection can now be updated after it is created, so changes made in the interface are saved correctly.
• Cloud Connections: OpenAI cost data now imports reliably for the largest connections that previously failed when a single cost response was too big to process at once.
• CloudZero MCP: When the assistant requests optimization recommendations with an unsupported filter or grouping, it now recovers and retries instead of failing your question outright.
• Dimension Studio: The cost source dimension now shows a readable label instead of its raw internal column name wherever dimensions are listed.
• Dimension Studio: Publishing a cost context that fails now returns a clear, actionable message explaining what to fix instead of a generic server error.
• Explorer: Hovering over the stacked line chart now highlights the band directly under your pointer, so the tooltip, highlight, and drill-in all match the section you are pointing at.
• Explorer: The cost-over-time chart legend no longer lists series with less than one cent of spend, so the key matches what the chart and table actually show.
• GitHub Connection: GitHub cost data now recovers automatically after a temporary export conflict, instead of reporting success while silently returning no new data.
• macOS Collector: On networks that advertise IPv6 without a working route, everyday web browsing no longer breaks while AI capture is running, so both keep working together.
• macOS Collector: The app now detects when an upgrade left an outdated network extension running and re-activates the current one, so your traffic is no longer handled by stale code.
• macOS Collector: Codex Desktop activity is now attributed to Codex instead of an unknown tool, so its usage and cost group with the rest of your Codex spend.
• Optimize: The insights header now shows correct total cost and status counts for organizations with more than 200 insights, instead of undercounting from a single page of results.
• Optimize: Paging through recommendations sorted by a column with many repeated values now returns every recommendation exactly once, with no duplicated or dropped rows.
• Snowflake Connection: Compute and cloud-services usage rows now display their unit as Credits; the underlying quantities and rates were already correct.
• Snowflake Connection: Archive storage and network charges that were previously left out of imported Snowflake costs now appear, so your Snowflake spend reflects the full bill.