Changelog
What’s new in Superbrain
Release notes for the beta. New builds reach testers automatically through in-app updates, no reinstall needed.
September 2026
- v0.7.1
- Added yearly billing for Prevue and Forge, with your allowance still renewing each month.
- Added switching a subscription between monthly and yearly billing.
- Added changing the email on your account through a support ticket.
- Added a short "What went wrong?" choice after a thumbs-down, usable from the keyboard.
- Changed Regenerate to sit in the response's … menu and to ask before replacing a long answer.
- Changed the thinking effort control to offer only the levels your plan includes, and to hide it when there is only one.
- Changed the code gathered for each message to favour the files you are working in, which finds the right code more often.
- Fixed an answer cut off by a dropped connection being shown as finished — it now says it was interrupted and offers Try again.
- Fixed Try again starting an interrupted task over — it now continues from where it stopped.
- Fixed Stop not stopping the model once a reply had started, which kept counting against your allowance.
- Fixed a reply you stop costing several times what it would have cost to finish.
- Fixed Flux failing partway through writing a large file.
- Fixed a slow Flux reply failing the whole turn — it is now retried in place.
- Fixed an OpenAI reply that hit its output limit being treated as finished.
- Fixed Delete on an earlier request removing it and every turn after it without asking.
- Fixed confirmation dialogs on macOS showing no question, only an explanation and a button.
- Fixed Copy on a response copying the whole work log instead of the answer.
- Fixed thumbs up and thumbs down on a response never being recorded.
- Fixed semantic code search timing out and returning nothing on about half of searches.
- Fixed a code search model download that dropped near the end starting over — downloads now resume.
- Fixed the code search model downloading several times over on a first launch.
- Fixed code indexing failing on every launch after a GPU stopped working — it now falls back to the CPU.
- Fixed code search still finding a renamed function under its old name.
- Fixed who-calls-what answers going stale as you edit.
- Fixed symbol details never including the symbol's documentation in Python and Rust.
- v0.7.0
- Added Flux, which chooses a model for each request on the server and moves to another provider when one is slow or unavailable.
- Added a thinking effort control to the model picker, so you can trade speed against depth on models that support it.
- Added replay of the previous turn's reasoning on models that ask for it, which keeps long pieces of work coherent across turns.
- Added image attachments on Flux, routed to a model that can read them.
- Added Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash and Gemini 3.5 Flash Lite.
- Added plans. Free, Prevue and Forge, each with its own usage allowance and model access.
- Added a usage meter to the account panel, with a warning before the allowance runs out and none when it will last.
- Added top-ups, so you can buy more usage part way through a period without changing plan.
- Added an all-time usage total to the account, which no renewal or plan change resets.
- Added error messages that carry a title, a plain description and only the actions your plan allows.
- Added a one-time notice when a plan change removes the model you had selected, naming the model and what replaced it.
- Added a usage dashboard on onesuperbrain.com covering your allowance, spend, purchases and receipts.
- Added in-product support tickets, with replies delivered in the dashboard rather than by email.
- Added a pricing page that reads live plan and rate information rather than fixed copy.
- Changed Auto Mode to Flux, which is now chosen on the server and shown under one published rate.
- Changed the model list to arrive from the server, so new models can appear without updating Superbrain.
- Changed the model picker to order models the way it groups them rather than alphabetically.
- Changed the account panel to separate your identity, your allowance and your keys into their own sections.
- Changed context window sizes to be published per model rather than fixed in the app, so a model's real window is what you get.
- Removed Gemini 2.5 Pro, Gemini 2.5 Flash, GPT-4o and GPT-4o mini, which their providers have retired.
- Fixed the Context Panel reporting $0.0000 and 0% full on a chat you reopened.
- Fixed session cost losing the work already done when you press Stop.
- Fixed session cost when a turn moved between models, each model is now priced at its own rate.
- Fixed thinking tokens being left out of the context estimate, which overstated how much room was left.
- Fixed stopping a turn cancelling an MCP server that was still starting in the background.
- Fixed Superbrain waiting longer than a turn should for an MCP server to report its tools.
- Fixed clearing your last API key leaving a model in the picker that could not run.
- Fixed an account menu action that deleted your API key.
- Fixed the account panel disappearing on macOS when a native title bar is in use.
- Fixed a provider error being reported as your sign-in having expired.
- Fixed a provider being taken out of service for ten minutes when there was nothing to fall back to.
- Fixed rate limit messages ignoring the wait the provider asked for.
- Fixed an error that suggested starting a new chat when the fix was to update Superbrain.
- Fixed image attachments being withheld without saying why, the pill now names the plan that allows them.
- Fixed a pause before the first reply when no folder is open.
- Fixed Superbrain asking the editor to activate an extension that was never installed.
- v0.6.1
- Added chat tabs — the side bar holds several chats at once, and gives them back when you reopen Superbrain.
- Added Generate Commit Message to the Source Control input, running on Superbrain's own models.
- Added a Co-authored-by trailer crediting Superbrain on commits that contain code it wrote — on by default, and turned off with
git.addAICoAuthor. - Added a command to move your MCP servers from
.vscode/mcp.jsoninto.superbrain/mcp.json. - Changed workspace MCP configuration to live in
.superbrain/mcp.json, with.vscode/mcp.jsonstill fully supported. - Fixed MCP servers delaying the first reply of every message — they now start in the background.
- Fixed a message waiting indefinitely when an MCP server never reported its tools.
- Fixed starting a new chat while one was still answering discarding it silently — Superbrain now asks first.
- Fixed Auto running a far more expensive model than intended whenever it could not run the one it was set to.
- v0.6.0
- Added support for MCP servers — connect one, locally or over remote HTTP, and the agent can use its tools alongside its own.
- Added interactive MCP App panels, which stay in view after the turn ends rather than folding away with the activity trail.
- Added Z.ai's GLM models as a provider (bring your own key).
- Changed semantic code search to narrow on the files you have open, which finds the right code substantially more often.
- Fixed Keep All doing nothing after the agent edited a file — the code was on disk but the diff bar never cleared.
- Fixed a diff showing the wrong before-image when two edits finished out of order.
- Fixed sub-agents being able to write outside your workspace, where the boundary check never ran at all.
- Fixed code decorations disappearing whenever an extension loaded a stylesheet from another origin.
- Fixed an agent file targeting GitHub Copilot being marked up with Superbrain's own rules.
- Fixed Sarvam and Kimi being unreachable on Auto, where every turn would have been refused.
- Fixed a general price row overriding a more specific one, mispricing models that match both.
August 2026
- v0.5.3
- Fixed semantic code search running out of memory and stopping on machines with 4–6 GB of RAM — its peak now fits in about 1.1 GB, down from roughly 3.2 GB.
- Changed the on-device code index to use about 3.6× less memory, with no change to result quality.
- Changed the code index to size its memory to your machine instead of assuming a fixed amount.
- Changed the search engine to release its memory after a spell with no searches, instead of holding it for the whole session.
- Removed the previous search model an update leaves behind, reclaiming about 1 GB of disk after upgrading.
- Fixed semantic search often missing the right code when a query pointed at a specific area of the project — it now finds it far more often.
- Sped up first-time indexing on multi-core machines by using every physical core.
- Fixed the code index being built twice at once when the engine starts.
- Fixed a search model that failed to load being deleted and re-downloaded in full on the next search — it is now left in place and retried.
- Added Xiaomi MiMo as a provider, with its chat models and pricing (bring your own key).
- Fixed a MiMo token-plan key returning a bare authentication error — it now says which key type to use instead.
- Fixed prompt-cache reuse being silently skipped on five of nine providers, so repeated turns now hit the cache and cost less.
- Fixed parallel sub-agents retrying in lockstep after a rate limit — their retries are now staggered so work recovers faster.
- Fixed a rate-limit retry occasionally waiting up to 12 seconds against its own 10-second cap.
- Fixed a provider's refusal to answer being treated as a completed response.
- Changed onboarding to ask which provider you want before asking for its key.
- Redesigned the first-run landing screen, which had been a wall of text with no clear next step.
- Fixed the model picker not updating when the default model changed while it was open.
- Fixed extension installation being blocked for every extension — installs work again.
- v0.5.2
- Fixed a file referenced by a path relative to your project being reported as not found, in both the single-file and batched reads.
- Changed a directory listing or search that would come back empty to recover what was meant and answer it in place, instead of returning nothing and spending a full model turn on it - nearly one tool call in ten returned nothing at all, and three in five directory listings.
- Fixed test output written to stderr being discarded whenever the command also succeeded, which hid passing runs.
- Fixed the detected command for running your project's test suite being trimmed out of the context it was added to.
- Fixed a Windows-only
cd /dreaching a POSIX shell, where it changes into a directory named /d and runs the rest of the line somewhere else - now caught on every run, including ones that report success. - Fixed pressing stop during a search being reported as "no matches" rather than as a cancellation.
- Changed a fetched web page to return up to 65,000 characters in one read instead of 3,500, so a long page no longer costs a model round-trip per section.
- Fixed the session cost total re-pricing finished turns at the hour it happened to re-render in, which showed up to double or half the real cost on a provider whose rate moves during the day.
- Fixed the hours between DeepSeek's two peak blocks being billed at the peak rate instead of off-peak.
- Fixed DeepSeek's move to time-of-day pricing being applied 16 hours before it took effect.
- Changed DeepSeek to bill off-peak all day at weekends, following Beijing's weekend rather than UTC's.
- Removed Opus 4.7 from the model picker - every request to it is rejected by the provider.
- v0.5.1
- Fixed the cheapest OpenAI model being charged at five times its published rate, and GPT-5.6 Terra at 25% over.
- Fixed a Sonnet price rise the provider later cancelled still being applied, overstating every Sonnet and Auto turn by 50%.
- Fixed a cache-write fee being charged on models that publish none, overstating cached turns on GPT-5.5, GPT-5.4-mini and GPT-5.3-codex by 25%.
- Fixed cache hits on Kimi being counted as fresh input, which also made Kimi look as though it never hit cache at all.
- Fixed the long-context price step above 272K tokens being missing on the GPT-5.6 line and GPT-5.5, where the whole request costs 2x input and 1.5x output.
- Fixed a Grok prompt of exactly 200,000 tokens billing at the short-context rate instead of the long one.
- Fixed Sarvam being priced at roughly a seventh of its published rate.
- Fixed Sarvam usage counting against your beta credit in rupees as though they were dollars — a 96x over-count that could exhaust a window in one or two turns.
- Fixed Auto being priced as one fixed model regardless of what it actually ran, overstating output by up to 2.5x when it routed elsewhere.
- Changed DeepSeek to bill on the peak and off-peak schedule it moved to on 16 August, instead of the flat rates it left behind.
- Changed your remaining beta credit to be metered against the same prices the cost display shows, so the two now agree to within a millionth of a dollar.
- Added Claude Opus 5 and Grok 4.6 to the model picker.
- Fixed the out-of-credit message offering your own API key as the only way to continue — it now names the time your free beta window renews.
- Added directory deletion, moved to your system trash rather than removed outright, so a corrupted build cache can be cleared without dropping to a shell.
- Fixed Superbrain refusing to create the files a request described whenever the wording mentioned an existing file as an example.
- Fixed commands that discard their output, such as a redirect to /dev/null, being blocked as though they wrote a file.
- v0.5.0
- Added the ability to read pages from the web during a task, opening a fetched page as an addressable document so Superbrain can reach the section it needs instead of only the first screenful.
- Added support for reading local addresses — localhost:3000, a machine on your LAN, an internal wiki — so "check what my dev server returns" now works.
- Fixed page-section selection recognising only Latin script, which returned page one for most of the world's documentation.
- Changed a rate-limited or briefly failing page to be retried within the same step, and never cached as a failure.
- Fixed a failed fetch reporting a generic error instead of the actual cause.
- Fixed redirects being followed past the safety checks meant to inspect them.
- Changed an under-filled excerpt to stay short rather than be padded with unrelated text.
- Changed "no errors found" so it can no longer mean "nothing was checked".
- Fixed Superbrain claiming a file had been analysed by the language service when it had not.
- Changed work to be verified by running something that exercises the change, rather than by reasoning about which files it touched.
- Fixed a failing test on code just changed being treated as unrelated noise instead of as evidence against the change.
- Changed environment problems to be ruled out before a failure is blamed on your code.
- Fixed a test suite that asserts the old behaviour being treated as proof the new behaviour is wrong.
- Changed the completion check to ask what a command can actually catch, not merely which command ran.
- Fixed a file with no testable units silently voiding the check that a change was covered.
- Added an opt-in check that answers "did my change cause this?" when something fails in files Superbrain never edited — off by default.
- Fixed quoted text in a request, such as "change the label to 'Save changes'", blocking the first edit and then searching for wording that existed only because you asked for it.
- Added a read-before-edit requirement to the editing flow, so a file is never changed on assumption.
- Fixed Superbrain being told it could read only 100 lines at a time when 2,000 were allowed, which cost extra round trips on every large file.
- Fixed a filename at the end of a sentence going unrecognised when scoping the work.
- Changed related tool calls to be issued together where possible, so multi-file work finishes in fewer steps.
- Changed an upstream fix to be adopted as the change itself rather than as a copy of the current file, so it matches the version of the code you have.
- Fixed long command output being cut before it could be budgeted or saved, and changed truncation to report the full amount withheld.
- Changed overflow from a truncated result to be kept and retrievable instead of discarded.
- Fixed the saved-output folder growing without limit.
- Fixed a read guard blocking verification commands that were never file reads.
- Fixed images and re-sent messages rebuilding the cached prompt from scratch, which quietly made follow-ups more expensive.
- Fixed a full context ledger being mistaken for an exhausted budget, ending tasks that still had room.
- Added a limit on how much a single batch of parallel tool calls can return, so one wide fan-out cannot flood the context.
- Changed the context panel to fold the visual row into Conversation for a clearer breakdown.
- Added a summary at startup of which capabilities are switched off, and why.
- Fixed adding your own API key not taking effect until restart.
- Fixed image handling claiming a quality tier that produced no usable text.
- v0.4.10
- Fixed long-running tasks stopping partway through and asking you to resend your message — Superbrain now renews your session in the background and the task keeps going.
- Changed session renewal to cover every tier, so Auto and your own API key behave the same on long tasks across every supported provider.
- Added automatic recovery from a dropped connection — Superbrain waits and retries for up to 5 minutes, then picks the task up exactly where it left off.
- Added a reconnecting notice while the connection is down, so a pause is never mistaken for a freeze.
- Changed brief provider outages, overloads, and rate limits to be waited out rather than ending the task.
- Changed a replaced API key to take effect on the next step of a running task, instead of after a restart.
- Changed Stop to take effect immediately while Superbrain is waiting to reconnect.
- Changed sign-in and billing messages to state what actually needs doing rather than asking you to try again.
- Changed cost reporting to price each provider's prompt cache at that provider's own rates, so the figure you see matches the bill you get.
- Fixed reasoning blocks and images counting as zero tokens in cost and context figures.
- Fixed cached context being dropped early on providers other than Anthropic, which quietly made follow-up questions cost more than they should.
- Fixed delegated work re-sending its shared setup on every sub-agent run.
- Changed
@fileattachments to be budgeted per file, and to say plainly when part of one was left out. - Changed citations to resolve to exactly one file or none — an ambiguous reference is now reported instead of guessed.
- Added a closing note listing any statement in an answer that isn't backed by a file Superbrain actually read.
- Fixed a citation pointing at a line Superbrain never read being presented as verified.
- Changed read tracking to carry across turns, so a file read earlier in a session still counts as evidence for a later answer.
- Fixed answers being treated as edit requests because the wording happened to contain a word like "update" or "fix".
- Fixed a symbol's name influencing what kind of request Superbrain thought you had made.
- Fixed paths differing only in letter case being treated as two files on case-insensitive filesystems.
- Added
superbrainandsbterminal commands — open any folder in Superbrain straight from your shell. - Added automatic setup of the terminal command on macOS at first run, without a password prompt where possible.
- Added the terminal commands to the Windows and Linux packages, so they are on your PATH right after install.
- Added shell completions for the terminal commands.
- v0.4.9
- Fixed broad, whole-codebase questions being worked one file at a time — the agent now delegates wide exploration to fast background sub-agents and answers from their summaries, keeping the chat quick and lean.
- Fixed the agent being limited to a single background exploration per session — it now runs several in parallel and revisits new areas as the task moves.
- Fixed background exploration not saying what it left uncovered, so the agent no longer over-claims or repeats searches it already ran.
- Fixed the debugging helper not reasoning through a failure before proposing a fix.
- Fixed code review routing the whole diff through the main chat — the review sub-agent now fetches it directly, keeping the conversation lean.
- Fixed a multi-second freeze when renaming a symbol or repeating an edit across a very large file.
- Fixed a hang when an edit's search text was empty, and confusing failures when two edits to the same file overlapped.
- v0.4.8
- Fixed edits to files with Windows line endings occasionally corrupting the edited line — replacements now keep the surrounding characters exactly intact.
- Fixed the agent stopping after five files when asked to edit more than five in one request; it now edits every file you name.
- Fixed whitespace-only edits, like adding a trailing blank line, being wrongly rejected as "no change."
- Fixed the agent sometimes editing the wrong part of a file on the first try, so more edits land correctly without a retry.
- Fixed the agent occasionally refusing to re-open a file it had already read during a long editing session.
- Fixed slow edits on very large files.
- v0.4.7
- Added free Auto on every surface — the chat panel, inline chat, and any extension that asks Superbrain for a model now run on your beta credits, with no API key required.
- Changed Auto to stay free even after you add your own API key — Auto runs on your beta credits and never spends your key.
- Added a usage meter to the account menu showing how much of your free window you've used and when it renews.
- Changed Superbrain to tell you exactly when your free window renews, and offer to continue on your own key, instead of failing quietly when it runs out.
- Changed the context panel to read "Auto Mode" and "Sub-agent" while on Auto, instead of showing the underlying model name.
- Fixed answers arriving all at once — replies now stream in as the model writes them, so the first words appear right away.
- Fixed Stop not taking effect while a request was waiting to retry, which could leave the button unresponsive for up to 45 seconds.
- Fixed reasoning appearing twice in some chat surfaces.
- Fixed sub-agents continuing to run after you dismissed them.
- Fixed TokenFold missing generator functions — the agent can now find and reason about
function*andasync function*code. - Changed TokenFold to surface better matches on large repositories, so the right code turns up more often.
- Fixed slow code search on large repositories — results now come back noticeably faster.
- Changed TokenFold to make a freshly opened project ready to search sooner.
- Changed TokenFold to use less memory while it works.
- Fixed the first question of a session being slow to start — sign-in, keys, and connection setup now happen while the window loads.
- Fixed Superbrain signing you out when several windows started at once.
- Fixed macOS builds never updating themselves — every Mac had been stuck on the version it was first installed with, and updates normally from this release onward.
- Changed TokenFold to use your Mac's performance cores, making its first pass over a large project several times faster.
- v0.4.6
- Added the model's reasoning to chat — it streams live while the model thinks, then folds away when the answer starts.
- Added a
superbrain.thinking.displaysetting to keep reasoning open after a turn or hide it entirely. - Changed chat to follow the work as it happens, while leaving you in control the moment you scroll up to read.
- Added a pinned header naming the question the current answer belongs to; clicking it takes you back there.
- Changed the chat layout to a flatter typographic hierarchy, replacing the boxes and rails that framed each part of a turn.
- Changed "Early Access" to "Beta" throughout the app.
- Fixed the macOS download being named VSCode instead of Superbrain.
- v0.4.5
- Added a macOS build for Apple Silicon.
- v0.4.4
- Fixed the agent editing a file it guessed at from a vague request — it now front-loads a file for editing only when you name it (a path, a symbol, or a backticked identifier) and otherwise finds the target itself.
- Fixed the agent working from a stale copy of a file that changed mid-task — after a formatter, a codegen or build step, a
git checkout, or your own save, it re-reads the current contents before editing. - Fixed the agent skipping its read-before-edit checks when a request scoped out one action ("don't fix the tests, just rename this function").
- Fixed the agent stalling or ending a task too early on long, multi-step work — it keeps going while it is making verified progress.
- Fixed internal diagnostic and rate-limit text leaking into replies — the agent now answers from the evidence it gathered.
- Fixed the agent cutting work short on non-default models and on long background runs.
- Fixed wasted cost from the agent re-reading files it already had open — it reuses what it has read and re-reads only when a file changed.
- Fixed the agent treating files it hadn't touched as already edited.
- Fixed codebase audits leaning on assumptions about which bug matters most — findings are now ranked by impact demonstrable from the code.
Loading earlier releases…
Showing 2 of 4 months of releases.
Superbrain is in beta · 39 releases since June 2026