Changelog

Alerts now catch a stolen or forgotten API key early on every plan, and keep alerting as spend keeps climbing.

• NewDefault alerts for every workspace, on every plan: a provider API key that had no spend for 30 days starts spending, any API key spends far more in an hour than its busiest recent hour, or daily spend runs well above normal. If you have not set up Slack, email or Discord alerts, they go to the workspace owners by email and to the notification bell.
• NewSpend caps for any API key, feature, model, customer or tag: for example, alert when any API key spends $50 in an hour, or any feature spends $200 in a day. One alert lists every key or feature over the cap.
• NewNew or dormant spender alerts: get told when an API key, model, workspace or region that has not spent anything in a while starts spending.
• NewAPI key alerts name the key and link to the Anthropic or OpenAI console page where you can revoke it.
• ImprovedEvery alert now alerts again when spend passes 2x, 10x and 100x its threshold in the same day or month, instead of only once. Critical alerts are clearly marked in Slack, Discord, email and the bell.
• ImprovedAlert history shows how much was spent by the end of the day or month, not just the amount when the alert first fired.
• FixedAn alert could occasionally be sent twice. Each alert is now sent once.

The coding telemetry installer now asks for your Cursor API key on a hidden prompt instead of telling you to put it in your shell.

• ImprovedWhen you set up Cursor cloud agents, the installer offers to take your Cursor API key on a hidden prompt to check and store the repository's SuperPenguin secret. The key is used only during that run and never saved, so it no longer ends up in your shell history. Press Enter to skip.

SuperPenguin can now warn you within about an hour when AI spend jumps unexpectedly, including from a leaked provider API key.

• NewUnexpected spend detection: every hour SuperPenguin compares each workflow, model and provider API key with its own recent history and alerts you through your usual Slack, email or Discord alerts when spend jumps far above normal. It also watches provider API keys used outside the SuperPenguin SDK, so a leaked key that starts spending is caught quickly.
• NewEach alert explains what changed and suggests a next step. An ongoing problem alerts you again only if it gets much worse, and you can acknowledge it or mute it for a day or a week from the Alerts page.
• NewSuperPenguin now watches every connected Anthropic and OpenAI API key for signs of a leak, for every organization. If you have no alert channel set up, we email your organization's owners. Turn on full workflow and model monitoring, choose how sensitive it is, and pick which channels it uses on the Alerts page.

Clicking a 10-minute block on the Live API usage chart now shows its breakdown.

• FixedOn the API usage page in Live mode, clicking a bar or point opened a breakdown that said "No data" and $0.00 for most 10-minute blocks. The breakdown now shows that block's spend, requests and every section, and its totals match the bar you clicked.

Claude Code cloud sessions now keep their transcripts, however many repositories they open.

• FixedSession transcripts from Claude Code cloud environments were missing for sessions that opened more than one repository. The setup script line now turns transcripts on for every session in the environment, and keeps itself up to date.

Cloud agent setup now checks Cursor and Claude Code separately.

• FixedThe installer no longer says a repository's cloud agents are set up when only Claude Code is. It offers just the missing part, updates out-of-date Cursor hooks, and lists exactly what is left and how to add it.
• ImprovedWith a Cursor API key, the installer checks that Cursor has your SuperPenguin key for the repository, and creates a new cloud key only when one is actually needed.

One command now sets up every AI coding tool, with no keys to copy by hand.

• NewRunning the installer again is always safe: it checks Claude Code, Codex, Cursor and Gemini CLI, repairs anything that changed, and keeps your machine signed in.
• NewWhen your organization turns on session transcripts, the installer sets them up on its next run and tells you in one line. You can opt out on your machine with --no-transcripts.
• NewRun inside a repository, the installer offers to set up its Cursor cloud agents and Claude Code cloud environments, creates their key in the same browser sign-in, and shows exactly where to paste it. With a Cursor API key, it stores the key in Cursor for you.
• NewA new docs page, Track AI coding tools, explains the command, what it changes on each tool, and what is and is not sent.

Cursor activity and session transcripts, like Claude Code and Codex.

• NewCursor now reports each tool call the agent makes, such as terminal commands, file edits, searches and MCP tools, with how long it took and whether it failed. Only the tool's name and result are sent, never its input, output or file contents.
• NewCursor also reports when a subagent finishes and how long each session lasted, so you can see which agent turns did the most work.
• NewIf your organization has turned on session transcripts, the installer now uploads Cursor conversation transcripts too, with the same opt-in, retention and deletion as Claude Code and Codex.
• ImprovedYour Cursor spend still comes only from Cursor's billing; the new activity reports never change it.

The coding-tool installer now sets up Cursor too.

• NewThe one-line installer now sets up Cursor on your machine, in both the editor and the Cursor command line. After each agent turn, Cursor reports the model and the tokens it used to SuperPenguin, without prompts or responses.
• NewSettings > AI coding has a command that adds the same reporting to a repository, so Cursor cloud agents report their turns too.
• ImprovedYour Cursor spend still comes from Cursor's billing, so these per-turn reports never count the same usage twice.

• FixedSQL queries over individual requests that cover exactly 31 days with an inclusive end date, such as BETWEEN, are no longer rejected as too long.

SQL queries in API Usage, Transactions now follow the dates in your query.

• FixedThe date picker no longer limits SQL queries. The dates in your WHERE filter decide what is read, and the editor shows the period each query covers.
• FixedStarter queries now run within every plan's per-query limit. Totals by day, month and customer read the hourly summary, and examples over individual requests or tags look at one day.
• ImprovedThe SQL editor shows only the starter queries and tables your plan includes, and lists the columns of each table. Growth plans get the hourly usage and tag tables; Pro adds individual requests.
• ImprovedWhen a query is too large, the message shows how much it would read against your limit and how to narrow it.

Query and chart your API transactions with SQL, on a page that fits your screen.

• NewThe SQL view in API Usage, Transactions now offers starter queries: daily usage by model, monthly totals, one day hour by hour, top customers by cost, and the latest requests.
• NewQuery results can be shown as a chart. Results by date or time become stacked bars you can split by model, provider or another column, and other results become a ranking.
• ImprovedRows and SQL now share one toolbar, your query and results stay put when you switch between them, and the table or chart fills the rest of the screen without scrolling the page.
• ImprovedEvery column you can query is listed with a short description, grouped by purpose, and the list can be hidden to give your query more room. Press Cmd+Enter or Ctrl+Enter to run.
• ImprovedLarge results stay responsive. The page keeps up to 5,000 rows and charts the most recent 200 time periods, and says so when it stops.

• ImprovedWhen several optimization recommendations are published for your organization at once, owners now get one email that lists them all with their combined estimated monthly savings, instead of one email per recommendation.

Set up coding-tool telemetry once for your whole team.

• NewOrganization owners can create one organization key in Settings and paste ready-made blocks into Claude Code managed settings and Codex managed config, so every developer is covered without installing anything on their machine.
• NewUsage sent with the organization key is credited to the teammate whose Claude or ChatGPT account email matches their SuperPenguin email. Emails that match no one are shown as unmatched.

• FixedClicking a chart no longer leaves a grey border around it. Keyboard users still see a focus ring when they tab to a chart.

Organizations can choose to keep full coding session transcripts.

• NewCoding transcripts (opt-in, off by default): organization owners can choose to store full Claude Code and Codex session transcripts for 90 days.

The date picker lets you view any past month and step between months in one click.

• NewArrows beside the date picker step back and forward one whole month, so you can compare two or three months ago without opening the calendar.
• NewPick any earlier month from a month grid, not just this month or last month.
• ImprovedCustom ranges have their own option with clear start and end dates and a preview of the range as you choose it.
• ImprovedThe month-over-month comparison follows whichever past month you select.

A notification bell shows spend alerts and new recommendations.

• NewA bell at the top right of the dashboard lists recent spend alerts and new optimization recommendations, with a badge for anything you have not seen yet.
• NewSpend alerts now appear in the dashboard even if you have not connected Slack, email or Discord.
• NewWorkspace owners get an email when a new optimization recommendation is published for them.

Optimization ticket activity always finishes loading.

• FixedSwitching optimization tickets while a comment is still saving no longer leaves the new ticket's activity stuck loading.

Track reviewed savings recommendations as tickets in Optimizations.

• NewOptimizations now shows reviewed recommendations as tickets with estimated monthly savings, the evidence behind each one and the recommended steps.
• NewAssign an owner, mark a ticket complete, dismiss it with a reason or reopen it, and discuss the plan with your team in comments.
• NewCopy a ready-made prompt from a ticket into your coding agent to start the fix.

API Usage bar tooltips show the usage you hover.

• FixedAPI Usage bar charts show the correct date and value when earlier time periods have no usage.

API Usage opens using your saved timezone.

• FixedAPI Usage waits for your saved timezone before loading usage data.

Connect your tools first, then add provider billing when you are ready.

• ImprovedOnboarding separates connecting your tools from optional provider billing, making it easier to start tracking costs.
• ImprovedInvited teammates can connect tools and optionally add billing without repeating workspace setup or choosing a plan.
• ImprovedClearer setup instructions and more readable onboarding screens help you get started on laptops and smaller screens.

Compare AI coding costs across people and teams, with easier organization setup.

• NewExplore AI coding costs by person and team, switch between actual cost and usage value, and download the comparison as a CSV.
• NewImport your organization structure from a CSV or pasted spreadsheet cells, review reporting lines, and update people in Settings.
• ImprovedPeople with no recorded AI coding usage remain visible, with clearer explanations when usage is unavailable or not reporting.
• FixedTogether and Anthropic pricing updates preserve free rates and distinguish prices for different context lengths more accurately.
• FixedThe demo booking form loads reliably and stays readable on smaller screens.

API Usage averages remain available for historical usage.

• FixedAPI Usage shows average cost per request again and removes the cost-completeness warning.

Track agent workflows and Cohere usage, with clearer live spending and team setup.

• NewTrack model usage across agent workflows and attribute each run to its customer, team and feature.
• NewTrack Cohere chat, embeddings and reranking, including supported AWS Bedrock calls. Requests without reported billing usage keep an unknown cost.
• NewAPI Usage adds live spending charts with line and stacked-bar views.
• ImprovedInvite teammates with control over who can connect providers, and keep invited users in the intended workspace.
• ImprovedFollow clearer AWS and Azure setup guides, and use the SuperPenguin CLI to inspect your usage.

Easier-to-read PR attribution and more accurate model pricing.

• NewPR attribution shows how many requests are on record for each pull request and the value of each request, including requests without a known price.
• ImprovedPR attribution shows teammate names instead of anonymous IDs, keeps each pull request on one readable row, and shows whether it is open, merged or closed with an icon beside its number.
• ImprovedPR session details lead with the teammate, tool, model and cost, with token details grouped under Usage details.
• FixedGemini models used through AWS Bedrock and more Cursor model names, including Fast mode, are now priced correctly.

Clearer PR details and more consistent model pricing offers.

• ImprovedPR attribution now explains when model, request, token or session details are unavailable in saved records, without assuming the original tool omitted them.
• FixedModel pricing pages avoid duplicate offers for the same provider route while preserving historical prices.

Setup is easier to finish, and coding spend is easier to understand.

• NewA refreshed setup flow makes it easier to connect tools, invite teammates and choose a plan, with a 14-day trial for eligible paid plans.
• ImprovedSetup reminders follow your timezone and help you pick up where you left off.
• ImprovedPR attribution makes available session details easier to browse and explains when historical information or counts are unavailable.
• FixedOpenRouter spend reporting better reflects cached tokens, reported charges and usage with your own provider keys.

High-volume data capture is more reliable across distributed applications.

• FixedFixed missing captured request details when many clients share a connection.

Model prices are easier to compare, and cached-token spend is more accurate.

• NewModel pages distinguish Fast offers from hosting options and show newer releases first.
• ImprovedOpenRouter model comparisons better reflect the lowest available hosting price.
• FixedCached-token usage is priced more accurately in spend reporting.

High-volume SDK usage now reaches spend reporting reliably across distributed applications.

• FixedFixed missing usage for teams that send SDK telemetry from many distributed clients through one API key.

Setup is easier to finish, and you can connect coding tools even when the emails do not match.

• NewOnboarding now keeps API setup steps in the checklist, adds coding integrations, and puts AI help next to those tasks.
• NewYou can link Cursor, Claude Code, and Codex to SuperPenguin when those accounts use a different email than the one you signed in with.
• ImprovedSuperPenguin Desktop stays usable while it scans local activity, and Settings can show your Cursor account in the installed app.

Overview shows Total usage as soon as it is ready, then fills in the month forecast.

• ImprovedOverview now opens Total usage first. The month forecast appears a moment later, and setup steps stay out of the way if you already have traffic.

API Usage shows your spend as soon as it is ready, then fills in the month forecast.

• ImprovedAPI Usage now opens the charts first. The month forecast appears a moment later, and setup steps stay out of the way if you already have traffic.

API Usage opens faster on accounts with a lot of request history.

• ImprovedAPI Usage now loads much faster for teams with heavy request volume. Spend and request totals stay the same.

Anthropic Claude Fable 5.1 usage now resolves with its current token and cache-read pricing.

• ImprovedClaude Fable 5.1 usage is recognized with the correct input, output, and cache-read rates.

AI Coding attribution stays current as teams grow, API Usage avoids duplicate request totals, and new comparison guides make AI cost tools easier to evaluate.

• ImprovedCursor work reaches PR Costs through smaller, targeted updates, keeping pull request totals responsive as your team's coding history grows.
• FixedAPI Usage now counts a request once when the same delivery arrives more than once, preventing duplicate rows from inflating totals.
• NewThe new Library compares AI coding tools, observability platforms, gateways, and cloud cost products with practical cost-management guidance.
• ImprovedSuperPenguin Desktop 0.13.6 improves Claude Code coverage recovery and trusts current usage-limit readings sooner.
• ImprovedModel pricing recognizes more AWS Bedrock regions and Azure deployment variants so regional usage can resolve to the right rate.

Custom pricing now supports provider service tiers, and Sail batch usage can use the right price for each completion window.

• NewCustom model pricing now lets you add provider-defined service tiers with separate input, output, and cache-read prices.
• FixedSail batch usage now appears under Sail and can be priced independently for balanced and flex completion windows.

Overview now shows Total Usage across billed and SDK-attributed spend, AI Coding and PR Costs are easier to scan, and SuperPenguin Desktop 0.13.5 keeps the menu bar Day Sheet from interrupting your work.

• ImprovedOverview now reports Total Usage, so SDK-attributed spend (including providers without a billing connection) appears in the headline and by-provider chart alongside billed amounts.
• ImprovedAPI Usage can switch between the dashboard and a transactions list, PR Costs puts view filters in a bar above the table, and tabs show a matching layout while they load.
• ImprovedAI Coding makes Codex sessions easier to share and review, with a copyable session link, a clearer utilization list, and more helpful empty states.
• NewBook a demo is now available next to Start free on the marketing site, with a dedicated booking page.
• ImprovedSuperPenguin Desktop 0.13.5 is now available to install from inside the app. The menu bar Day Sheet opens without taking over your current window, Claude Code coverage stays accurate when you work offline, and Session Analyze matches the Sessions total.

A new SDK-first setup and dashboard experience gets you from signup to attributed AI spend faster, with clearer usage, reconciliation, and pull request views.

• NewNew workspaces now get guided setup with a copyable API key, instructions tailored to their stack, and live confirmation when attributed traffic starts arriving.
• ImprovedThe dashboard now uses a cleaner, collapsible workspace centered on API Usage, AI Coding, billing reconciliation, and provider details.
• ImprovedAPI Usage now forecasts the current month's SDK-observed spend instead of mixing in imported provider bills.
• ImprovedPR Costs now uses a denser, sortable table with dedicated views, clearer cost comparisons, and accessible row expansion.

SuperPenguin JS 0.8.0 and Python 0.13.0 record Vercel AI Gateway and LiteLLM billed spend, and Azure wrap docs cover regional pricing.

• ImprovedInstall SuperPenguin JavaScript 0.8.0 or Python 0.13.0 so Vercel AI Gateway and LiteLLM spend in SuperPenguin matches what those platforms charged for each request, including BYOK traffic.
• ImprovedAzure OpenAI wrap docs now show how to pass region and deployment type so SuperPenguin uses regional catalog prices instead of Global Standard.

Vercel AI Gateway and LiteLLM spend now matches what those platforms charged for each request, and Bedrock calls record the AWS region from the endpoint.

• NewCalls through Vercel AI Gateway or LiteLLM now use that platform's own per-request charge in SuperPenguin, including BYOK traffic, instead of SuperPenguin guessing from a rate card. This takes effect with the next SuperPenguin SDK release; current SDKs keep using SuperPenguin's rate card.
• FixedAWS Bedrock calls made through an OpenAI-compatible endpoint now record the AWS region from the host, so regional prices and region breakdowns are correct. No SDK upgrade is required for this fix.

AI Coding and PR Costs now guide your team through setup in one place, while Ask SuperPenguin can answer plan and company onboarding questions with workspace-aware guidance.

• NewAI Coding and PR Costs now share a live setup checklist for plan access, team invitations, Desktop connections, optional Cursor billing, and the extra GitHub connection needed for pull request attribution.
• NewAsk SuperPenguin can now compare plans and guide company setup, engineer invitations, Desktop installation, AI Coding access, and GitHub connection using the relevant status of your current workspace.
• ImprovedNew company setup guides explain workspace roles, plan limits, AI Coding visibility, and how to connect every engineer without requiring provider billing connections for Claude Code or Codex.
• ImprovedDesktop cost estimates refresh when SuperPenguin adds a new model price, instead of keeping a stale local cache.

AI Coding now compares estimated usage value by engineer across Cursor, Claude Code, and Codex, with clearer desktop invitations and more resilient pull request tracking.

• NewThe AI Coding overview now shows the average estimated usage value per reporting engineer across Cursor, Claude Code, and Codex, plus a ranked breakdown of the highest-usage engineers.
• ImprovedDesktop install invitations now explain that the app tracks costs and remaining usage limits across all three AI coding tools, and the email stays readable in dark-mode inboxes.
• ImprovedSuperPenguin Desktop 0.13.3 is now available to install from inside the app, with more dependable usage-limit reporting and fewer interruptions while you work.
• FixedFixed recent pull requests remaining missing after an interrupted repository refresh recovered. PR Costs now catches up immediately instead of waiting for a later refresh.

Cost per pull request now covers the Cursor work you do in your editor, not just cloud agent runs, and the desktop usage limits panel reports Codex, Claude Code, and Cursor again.

• NewPull requests now show the cost of Cursor work done in your editor. Each match is rated by how it was established, from the branch you were working on down to a model comparing what the conversation changed against the pull request, so most editor sessions now reach the pull request they contributed to.
• ImprovedThe share of Cursor spend attributed to pull requests is now measured against your full Cursor bill. It previously left out spend that could not be matched, which made coverage look higher than it was.
• FixedFixed pull requests going missing or stale on the PR Costs page. New and updated pull requests are picked up again.
• FixedThe desktop usage limits panel reports Codex, Claude Code, and Cursor limits again, and says why when a limit genuinely cannot be read instead of showing an empty or zeroed plan. Reading Claude Code usage no longer pushes Claude Code into its own rate limit, and SuperPenguin no longer takes keyboard focus while Claude Code is running.
• ImprovedSharing your AI coding usage limits with your company now asks once instead of repeating the prompt, and picks up an admin policy change right away.
• ImprovedPR Costs shows a page-shaped placeholder while it loads instead of a blank area, and the table no longer shows an internal label next to pull requests that have cost. Dashboard cards also carry a little more depth, so sections are easier to pick out.
• FixedCorrected Deepgram streaming rates in the pricing catalog so transcription spend is priced accurately.

Cost per pull request now counts the work your Cursor cloud agents did, and Cursor spend that cannot be tied to a pull request is shown plainly instead of quietly left out.

• NewPull requests now show the cost of the Cursor cloud agent work that produced them. Previously almost none of this reached the page, so pull requests that an agent wrote could look free.
• ImprovedCursor spend that cannot be confidently tied to a specific pull request is now listed as unmatched rather than dropped. The page adds up to what you were billed, and you can see how much is still unaccounted for instead of guessing.

Cost per pull request is now on for every Pro and Enterprise team automatically, and a new version of SuperPenguin Desktop is ready to install.

• NewCost per pull request now turns on by itself for every team on Pro or above. If your plan includes it, connect GitHub and it starts working, with nothing to request and nobody to wait on.
• NewSuperPenguin Desktop 0.13.1 is ready to install from inside the app. It separates spending included in your plan from spending beyond it using the same wording across Cursor, Claude Code, and Codex, rebuilds the coverage tiles to show whether usage came from a subscription or from credits, and reads Claude Code limits more often as an allowance runs low.
• ImprovedPull request costs settle faster during heavy activity. When several coding sessions finish at once, the affected pull requests now update in one pass instead of recalculating repeatedly.

PR cost attribution now works for every team on a Pro plan or above, starting the first time you connect GitHub.

• NewAny team on Pro or above can now see cost per pull request. Connect GitHub and start coding, and your pull requests fill in on their own with no setup step and no waiting for a historical import.
• ImprovedConnecting GitHub is more reliable. The first scan of your repositories now resumes on its own if it is interrupted, so a large account no longer ends up with an empty pull request list.
• ImprovedPull request costs update faster after you push. Only the pull requests you actually changed are recalculated.
• ImprovedAI Coding now uses the same within-plan and beyond-plan wording across Cursor, Claude Code, and Codex, so plan usage is easier to compare across tools.
• FixedFixed Claude Code usage being left out of plan coverage for sessions run outside the app, and for sessions that had no clear end.
• FixedFixed a Cursor model with no published price being counted as though it had been priced.

Card-free trials now give teams a clear path to evaluate SuperPenguin before choosing a paid plan.

• NewEligible teams can activate a Growth, Pro, or Enterprise trial without adding a payment card.
• ImprovedNew-team trials begin on first sign-in, so setup time does not reduce the evaluation period.
• ImprovedBilling now shows the active trial tier, end date, time remaining, and the plan that follows the trial.

PR Costs now connects more of the work behind each pull request and makes unmatched Cursor spend visible.

• NewPR Costs can now connect multiple coding sessions to one pull request and split a session across multiple pull requests, giving teams a fuller view of AI-assisted development cost.
• ImprovedDirect pull request creation and later contributions are recognized across Cursor, Claude Code, and Codex, so work is less likely to be missed when it spans branches or sessions.
• ImprovedCursor billing is now shown as attributed and unclassified spend together, so the page reconciles to the billed total instead of hiding the portion that has not been matched.
• ImprovedSuperPenguin Desktop now captures richer work structure for pull request matching while keeping prompt collection controlled by each person's privacy setting.
• FixedPull requests matched only to Claude Code or Codex now open correctly in the detail view.

A rebuilt superpenguin.ai, pull request costs for Claude Code and Codex, and a SuperPenguin Desktop that can explain your spend.

• NewSuperPenguin Desktop can now analyze a session and explain where its cost came from and what to change, without leaving the menu bar.
• NewPull request costs now include Claude Code and Codex work alongside Cursor, so a pull request shows what it cost across every coding tool your team uses.
• NewSettings now let each person choose whether their coding prompts are stored. Prompt storage is off by default, and turning it off does not stop pull request matching.
• ImprovedSuperPenguin Desktop signs you in through your browser and keeps you signed in across restarts, and it now shows how much of your Cursor plan allowance you have used this period.
• ImprovedAPI Keys has a redesigned page at a clearer address, and Attribution is now labelled SDK Attribution so it is easier to tell apart from AI Coding.
• Improvedsuperpenguin.ai has been redesigned, with new pages covering attribution, coding ROI, and pricing, and model pages now cite a source for every published specification.

Clearer project and session costs, more reliable pull request matching, and steadier live spend in SuperPenguin Desktop.

• NewSessions now break down input, output, cache read, and cache write tokens, with a clearer explanation of how each session’s cost was calculated.
• ImprovedPull request costs now match current Cursor conversations more reliably and refresh after GitHub or desktop data changes.
• ImprovedProjects stay visible in quiet date ranges, and SuperPenguin Desktop now explains how conversations are grouped into projects.
• FixedGitHub connections now report missing pull request access instead of appearing to sync successfully with incomplete data.
• FixedDesktop live spend stays steady while you work and settles cleanly as billed usage arrives, without stale peaks or sudden drops.

Custom model prices now follow the correct provider on more OpenAI-compatible endpoints.

• FixedFixed Amazon Bedrock requests appearing as OpenAI, so provider attribution and custom Bedrock prices apply correctly.
• FixedFixed Cloudflare-hosted models appearing under their model maker and corrected affected historical custom-priced totals.

Cursor spend now adds up to the same number wherever you look at it.

• FixedFixed Cursor spend that could be undercounted. Some usage went missing from your totals after it had already been recorded, which made spend look lower than it really was. Past months have been restored and the totals now stay complete.
• FixedSuperPenguin Desktop no longer adds an estimate on top of hours that have already been billed, so the amount it shows lines up with the web dashboard instead of running high.
• ImprovedDesktop estimates now cover only the most recent activity that has not been billed yet, and they clear out as soon as the real amount arrives.

Claude Code and Codex join Cursor in AI Coding, with clearer spend coverage and a redesigned API Keys page.

• NewAI Coding now includes Claude Code and Codex alongside Cursor, so you can see sessions, projects, and spend across the tools your team actually uses.
• NewSuperPenguin Desktop shows one Sessions list for Cursor, Claude Code, and Codex, with a tool badge and a dollar amount on every session.
• NewOverview and Projects on Desktop filter by All, Cursor, Claude Code, or Codex, and show what a subscription or credits covered versus what you should expect to be billed.
• ImprovedPull request costs can break out Claude Code and Codex spend next to Cursor, with a clear label for each tool.
• ImprovedThe API Keys page is clearer: rename and revoke without clutter from the top bar, and Escape cancels a rename in progress.
• ImprovedThe Mac menu bar Day Sheet is faster to open and easier to dismiss, with fresher local spend while you work.
• FixedDesktop no longer shows your whole team’s Cursor spend when it cannot tell which Cursor account this Mac uses.
• FixedSlack spend alerts no longer repeat the same daily spike link, and monthly alert labels use a consistent calendar month.

Clearer AI Coding Insights from the desktop app, easier teammate setup, and fresher Cursor billing.

• FixedAI Coding Insights now shows activity from SuperPenguin Desktop even when local environment details were not uploaded.
• NewInvite teammates to install SuperPenguin Desktop, and see who on your team is already connected.
• ImprovedCursor billing updates appear sooner in AI Coding and related spend views.
• ImprovedAI Coding and PR Costs setup now walk you through GitHub and the desktop app, without skill-file steps.

Spend tracking stays reliable during brief planned maintenance.

• ImprovedYour SDK usage continues to record correctly even if our database is briefly unavailable for maintenance.
• ImprovedPricing stays available for new traffic during those windows so costs still calculate.

More accurate Local-timezone spend, plus a Mac desktop download on the site.

• FixedSpend views in Local timezone now match across Breakdown and provider totals.
• FixedFixed date-range and month-to-date edge cases that could overcount or undercount spend.
• NewAdded a Mac desktop app download on the marketing site.
• ImprovedCleaner public pages and outbound links for search and sharing.

Smoother onboarding, public status and security pages, and more model pricing.

• NewNew public System Status page so you can check SuperPenguin availability.
• NewNew public Security overview covering how we protect your data.
• NewExpanded onboarding with more integration options.
• ImprovedRefreshed SuperPenguin Desktop look and in-app updates.
• ImprovedMore models in pricing, with updated public model reference pages.
• ImprovedMore accurate Cursor PR cost matching.

Better spend forecasts, multi-channel alerts, and more reliable signup.

• FixedFixed month forecast calculations on the Spend overview.
• ImprovedAlert notifications work across Slack, Discord, and email.
• FixedInvite and account activation work more reliably during signup.

• NewAdded Grok 4.5 long-context pricing and a public model reference page.

• ImprovedExpanded public model pages for Kimi K3, Mistral, MiniMax, and xAI.
• FixedAzure cost imports are more reliable when the provider is busy.
• ImprovedOpenRouter pricing now reflects flex, priority, and regional tiers correctly.

• ImprovedClearer signup progress tracking on the marketing site.
• FixedSignup started and completed events now stay linked to the same person.

More models in pricing and on public reference pages.

• NewNew public model pages for Inkling, Muse Spark, and selected Bedrock and Azure models.
• ImprovedExpanded pricing for Meta, Tinker, and OpenRouter upstream tiers.

• ImprovedSuperPenguin Desktop: flexible time ranges and faster charts.
• ImprovedPublic model pages are easier for search engines to understand.

SuperPenguin Desktop for macOS and richer AI Coding insights.

• NewLaunched SuperPenguin Desktop for macOS with device sign-in and live Cursor spend.
• NewAdded AI Coding Local Lab with per-project spend breakdowns.
• ImprovedAdded GPT-5.6 pricing and better OpenAI cache-write cost capture in the SDK.
• ImprovedExpanded the public model catalog with fuller reference details.

• ImprovedContent capture sampling adapts automatically for high-volume traffic.
• FixedFixed a demo banner display glitch during provider import.

• ImprovedConnect sheet shows live progress while importing provider history.
• FixedFixed cases where spend attribution could undercount large accounts.
• ImprovedImport-health notices match your plan's history window.