xAI Release Notes

Describe a job once and Grok runs it on a schedule or when an email arrives, then reports back.

Today, we're introducing Automations: jobs Grok runs on its own.

Describe the work once, choose when it runs, and Grok takes it from there, whether that's research done before you're awake or an important email flagged the moment it lands. Automations are available on grok.com and in the Grok app on iOS and Android.

Describe the job once

An automation's instructions read like any chat message: describe what you want, attach files for context, add connectors and skills, and pick a mode. Name it, save it, and from then on every run is a fresh request: same instructions, current data.

On a schedule, or on a trigger

Schedules run once, daily, weekdays, weekly, monthly, or yearly, at a time you choose in your timezone: a morning brief at 8:00 before the day starts, or a rent reminder on the 1st.

Email triggers watch your inbox instead. When an incoming email matches your filters (sender, recipient, or subject), the automation fires with that email as context, and Grok responds to the actual message. Instructions can also point at your tools directly: type @ to mention a connector, and Grok uses it on every run.

You can also start a run yourself with Run now, which is handy for testing an automation right after you build it.

Every run is a full conversation

When an automation fires, Grok opens a real conversation, does the work, and saves the result to its run history. Open any run to read the full thread, or pick up the conversation where Grok left off. You choose how each automation reports back: email, app notification, both, or neither if you'd rather check in yourself.

Start from chat, or a template

You can also create automations straight from chat: ask Grok to "check the news every morning and flag anything about pricing" and it sets one up for you. The Automations page has suggested templates to start from, and any automation can be paused, resumed, edited, or deleted at any time.

Get started

Create your first automation at grok.com/automations. Scheduled automations are available to everyone; email triggers are included with SuperGrok.

Explore the harness behind our coding agent and TUI.

We're open-sourcing Grok Build, SpaceXAI's coding agent and TUI. The source is now available on GitHub.

Publishing the code is the most direct way to build toward a robust and reliable harness. You can read the source to see exactly how it works, from context assembly to tool-call dispatch.

Open-sourcing also makes the harness easier to explore and extend: if you're working with skills, plugins, hooks, MCP servers, or subagents, the source is the definitive reference for how each is loaded and invoked.

Finally, Grok Build can now run fully local-first: compile it yourself, point it at your own local inference, and drive everything from your config.toml.

About the codebase

About the codebase

The published source includes:

• The agent loop: how context is assembled, how model responses are parsed, and how tool calls are dispatched
• The tools: how the agent reads, edits, and searches code, and how it runs commands
• The terminal UI: rendering, input handling, plan review, and the inline diff viewer
• The extension system: skills, plugins, hooks, MCP servers, and subagents

Explore the source on GitHub.

View on GitHub Browse the source on GitHub.

Get Grok Build Install with one command and run it in your terminal.

grok inspect now shows effective compatibility settings for Cursor, Claude, and Codex sessions

New setting "Match display refresh rate" lets high-refresh displays run the TUI at native cadence

Parked subagent status no longer duplicates or interleaves incorrectly in scrollback

Status line during waits now shows elapsed time before the queued-message hint

Queued messages sent with Enter now appear immediately instead of vanishing briefly

Resume hint after quitting minimal mode now prints the correct grok --minimal --resume command

Rate-limit messages now correctly direct API-key users to team plans instead of personal upgrades

Session picker now resumes recent Claude Code, Codex, and Cursor sessions

Welcome screen now offers a one-click resume nudge for recent Claude, Codex, or Cursor sessions

Web fetch tool preserves full truncated page content as readable artifacts instead of discarding it

Multiline mode now correctly sends the top queued message on empty Enter when a turn is running

Queued commands no longer disappear or delay when pressing Enter twice quickly during a running turn

Minimal mode text is now readable on dark terminals with proper contrast and highlighted user prompts

Grok no longer crashes when printing resume hints after the terminal pane has closed

Long-running turns with multiple waits now show updated status markers in the transcript instead of appearing stuck

Claude and Cursor hooks are now correctly disabled at session start when disabled in config

You can now pin authentication to API key or OIDC in config.toml; the unpinned method is no longer tried automatically

/context now shows token costs for skills and MCP servers

Linux middle-click paste from the primary selection now works; clipboard errors are handled more reliably

/terminal-setup now shows your terminal's color support level and which themes are available

grok setup --json prints your team's managed configuration without installing it

Messages you type while the model waits on tasks now stay queued; pressing Enter twice sends them immediately by cancelling the current turn

How-to Guides modal now shows a tip linking to Ask Grok above the footer shortcuts

Subagent task and spawn_subagent tools now accept an optional model parameter in the CLI

Keyboard Shortcuts modal now lists the paste key binding for images under the Input section

A pre_tool_use deny now feeds the reason back so the model can retry instead of cancelling the turn

Plan mode now strictly rejects edits outside the plan file, even under always-approve

Web search and X search no longer fail when both a local function tool and the backend hosted tool are active

SQLite databases no longer cause bus errors on network filesystems such as NFS

Resuming a session that is already open now focuses the existing view instead of creating duplicates

Turn completion markers in scrollback now read "Worked for …" instead of "Turn completed in …"

/btw loading spinner now animates correctly when the main session is idle

Mid-turn wait spinners now correctly show "Waiting on task output…" instead of Thinking

Scrollbar thumb is now visible in the oscura-midnight theme

Status messages for background work now end with a period

Editing queued prompts no longer freezes the terminal or duplicates text into the composer

Headless JSON output now includes token usage and cost per prompt and session

SDK turns now expose detailed token usage and cost information via Turn.usage

Double-click or Enter on a previous user message now lets you edit and resubmit it directly from the transcript

Text selection in scrollback now works better when starting on chrome, gaps, or while scrolling

Shell commands using rg (ripgrep) no longer require permission prompts by default

Voice mode is now available for API-key sessions

New environment variables allow tuning scroll and draw cadence for high-refresh displays

Background tasks started by the model in headless mode are now killed on exit instead of leaking

Agent process leaks on failed spawns and missing-stdio teardown are now prevented

Parked turn markers no longer appear after interjections and now count down as background tasks finish

The /context tool definitions line no longer shows the cryptic disclaimer suffix

Terminal release waits boundedly for the process to exit and repeated waits share the drain grace

Fixed a crash when the agent asked a question on narrow terminals

Fixed misleading keyboard hints in the /mcps panel in minimal mode

Clipboard copy now reports success correctly when using iTerm2 over SSH

Fixed scroll/input conflicts when plan approval appeared over an open edit block

Fixed frequent MCP and skills reloads that could freeze sessions on devboxes

MCP servers using HTTP (such as nebula for Slack) now automatically recover from disconnects

Next reset time in /usage now shows in your local timezone instead of Pacific Time

System notifications now carry structured kind/title/body for better rendering

x.ai/pr/status now reports whether an open PR is in the merge queue

Compact mode now activates automatically on very small terminals

Up arrow on an empty prompt now browses prompt history; /history searches it

Stop hook runs now appear inline on the turn-completed line instead of a separate block

Subagent rows now fold into verb-group headers and the tasks pane shows live activity labels

Dashboard shortcuts now advertise ? instead of Ctrl+. on terminals that cannot deliver the latter

Double-clicking scrollback while Text selection is fold/nav now shows a tip offering Ctrl+Y to enable Word select

grok worktree ls now works as a short alias for grok worktree list

MCP tool output truncation can now be set per-repo in .grok/config.toml

Auto-send of queued follow-ups during task waits can now be enabled fleet-wide via remote settings

Welcome screen now offers one-click resume of a recent Claude Code session via ctrl+u

Vim l key now opens the selected agent detail view in the dashboard

Terminal commands with no args now run through a shell, matching the CLI

Agent teardown no longer crashes on slim Linux images that lack the ps command

Esc now dismisses an open /btw panel before backing out of a dashboard overlay

Resumed grok.com chats now use the conversation's last model instead of the gateway default

JetBrains terminals on Windows now default to minimal mode to avoid raw mouse-report leaks in the prompt

Skill token highlights now survive line wraps and the slash menu opens when typing / before existing text

Truncated or tiny images are now dropped before sending and previously poisoned sessions self-heal on restart

Session switch hints after /new or fork now show the working command in minimal mode

Progress bars in Ghostty and WezTerm now stop correctly for parked task waits

/effort now rejects levels the current model does not support instead of sending a bad value to the API

/recap on a fresh session now says "No messages yet" instead of failing

Monitor and system messages no longer appear as user prompts when resuming old sessions

/rewind completion now appears as a brief toast instead of a permanent transcript line

Auto recaps no longer appear under a newer user message when you start typing again

Authentication retries after token refresh no longer hang for minutes or days

Text selection tip no longer appears on the first double-click or on non-assistant blocks

Skill slash commands queued while a turn runs can now be sent immediately with Enter or the interject chord

Drag text selection now works inside the dashboard dispatch input box

Multi-line paste chips in dashboard inputs now support preview and expand like the main prompt

Live previews now always fetch the latest content without browser or CDN caching

Release Notes

Teams can now ship default allowed commands via managed_config.toml (user deny rules still win)

Mid-turn interjections now appear as normal user prompts (❯) instead of a separate cyan block

IME text input in Otty no longer attaches unrelated clipboard images on every character

Rewind now fully removes the selected turn from both scrollback and the model's conversation history

Queued prompts now abort long blocking waits instead of waiting for the full timeout

File links and media now work for worktree sessions under ~/.grok/worktrees/

Collapsed Read/Edit tool rows now show only the filename instead of long absolute paths

Clipboard copies on Wayland now succeed even when the terminal loses focus mid-copy

User messages queued behind an auto-wake turn are no longer lost when the user presses Ctrl+C

Slash completion now shows sibling skills that share a frontmatter name and correctly sizes wrapped descriptions

Single tool calls that belong to a verb group now collapse into an aggregated header row

Fixed sessions that became permanently stuck after tool-use history corruption

/always-approve and /auto now toggle their mode on and off when run repeatedly

Terminal command cards on grok.com now correctly settle after foreground bash tasks

Copy failure toast now recommends trying /minimal for native terminal rendering

File watching on Linux now uses far fewer system resources for large projects with many dependencies

Changes

/sessions now opens the Agent Dashboard instead of a separate picker

New /goal slash command is now available

grok inspect now lists skills from [skills].paths and correctly labels bundled vs user skills

--minimal and --fullscreen choices are now remembered for future plain grok launches

Queued bash commands promoted at turn end now render their output instead of disappearing

Xcode / Foundation ACP clients can now drive grok agent stdio without silent parse drops on session/* calls

read_file now returns full single-line content (minified JSON, large dumps) instead of silently clipping at 2000 characters

Background task command preambles with newlines now render on separate lines instead of collapsing

Text selections now highlight uniformly even over inline code, links, and syntax-colored spans

grok --minimal now supports native drag-select on classic Windows conhost terminals

Skill tokens are now highlighted teal when used mid-sentence

Fixed a crash when a filtered list shrinks while the filter is active

Scroll lines and scroll speed settings now support fine unit-step adjustments

Project-specific plugins are no longer visible outside their project directory

First prompt no longer stalls for many seconds on large repositories while the filesystem watcher starts

Esc no longer cancels a running turn; use Ctrl+C instead. Double-Esc rewind now works while focused on scrollback

MCP permission prompts now show the planned arguments so you can judge what the tool will actually do

The "Managed by grok.com" link in the Extensions modal is now clickable and underlined

Dragging inside rendered markdown tables now selects whole cells or rectangular ranges and copies as TSV

Shift+Tab now goes straight to Plan mode when the plan-mode tip is showing

grok --minimal now aligns the prompt, status bar, and messages flush-left with the welcome card

/plugins no longer lists never-installed Claude marketplace entries and now groups plugins by their real source

Successful image compression no longer leaves a permanent line in the transcript

--no-ask-user now also disables ask_user_question for subagents

Fixed a crash shortly after launch on some systems caused by the telemetry exporter

• /minimal and /fullscreen commands let you switch the current session between minimal and fullscreen modes
• ask_user_question tool can now be enabled or disabled via config.toml, environment variables, or remote flags while defaulting to on
• User-run shell commands now display their complete output after finishing instead of silently dropping middle lines
• Edit tool output now correctly highlights multi-line strings and scopes that previously spilled across hunks
• Always allow grants for MCP, web_fetch and bash now take effect immediately in auto mode without re-prompting
• Cmd/Ctrl+click on bare http(s) links now opens only once on Warp terminals
• Cmd/Ctrl+click now works on imagine media paths and URLs that wrap across multiple terminal rows
• grok update on Windows no longer fails when a previous .old executable is still running
• Pasting images on macOS is now ~65× faster by reading the pasteboard directly instead of via osascript

Grok 4.5 is SpaceXAI's smartest model built for coding, agentic tasks, and knowledge work.

Today, we're launching Grok 4.5, SpaceXAI's smartest model built to excel at coding, agentic tasks, and knowledge work. It's our strongest model ever and was trained alongside Cursor.

Real-world engineering excellence

Grok 4.5 was trained on datasets spanning knowledge in coding, science, engineering, and math. With both intelligent and efficient reasoning, Grok 4.5 excels at real engineering tasks and exceeds comparable leading models at these tasks.

Benchmark bar charts of model scores. DeepSWE 1.0 (eval created by Datacurve, run with each model provider’s harnesses by AA): Fable (max) 66.1%, GPT 5.5 (xhigh) 64.31%, Grok 4.5 62.0%, Opus 4.8 (max) 55.75%, Opus 4.7 (max) 40.12%. DeepSWE 1.1 (mini-swe-agent harness run by Datacurve): Fable (max) 70%, GPT 5.5 (xhigh) 67%, Opus 4.8 (max) 59%, Grok 4.5 53%, GLM 5.2 44%. SWE Marathon resolution rate (pass@1): Grok 4.5 29.0%, Opus 4.8 (max) 26.0%, Fable (max) 24.0%, Opus 4.7 (max) 16.0%. Terminal Bench 2.1: Fable (max) 84.3%, GPT 5.5 (xhigh) 83.4%, Grok 4.5 83.3%, Opus 4.8 (max) 78.9%, Opus 4.7 (max) 78.9%. SWE Bench Pro resolve rate: Fable (max) 80.4%, Opus 4.8 (max) 69.2%, Grok 4.5 64.7%, Opus 4.7 (max) 64.3%, GLM 5.2 62.1%, GPT 5.5 (xhigh) 58.6%. Competitor figures are drawn from the respective developers’ published system cards or benchmark leaderboards.

Training Grok 4.5

Grok 4.5 was trained across tens of thousands of NVIDIA GB300 GPUs, with training and stability techniques designed for large-scale runs. Beyond raw token volume, we invested heavily in data filtering and curation: deduplication, quality scoring, and domain-focused selection so that the data mixture stayed high-coverage and high-signal.

We scaled reinforcement learning with a strong focus on per-token intelligence. Our RL training covers hundreds of thousands of tasks, centered on multi-step software engineering and other technical work, with automated and model-based grading. Our stack is built for highly asynchronous training, so agentic rollouts can run for many hours while learning continues across tens of thousands of GPUs. The result is more intelligent and efficient reasoning on real engineering and agentic tasks.

Built with one prompt

Grok 4.5 is incredibly capable at coding, from challenging Rust and C/C++ tasks to end-to-end app building from prompt to production. Below are some examples built by the model with one prompt. Grok 4.5 is highly proficient at creating well-designed, end-to-end functional apps even with minimal specification.

Faster than flash models

Grok 4.5 is served at fast-model speeds of 80 TPS. Combined with twice greater token efficiency than the latest leading models at the same tasks, the model delivers intelligent results to you more quickly and at far lower costs.

Token efficiency, average output tokens per SWE Bench Pro task — Grok 4.5 resolves tasks with 15,954 output tokens on average, about 4.2× fewer than Opus 4.8 (max) at 67,020

Excels at Office work

Grok 4.5 is now the default model in Grok Build. In addition to its coding proficiency, Grok Build is capable of building complex Excel models that involve research from the web, multi-sheet formula use, and even leaves stickies or notes behind for future reference.

In PowerPoint and Word, Grok 4.5 is similarly meticulous. The model is capable of using native PowerPoint shapes to build complex diagrams, designing intuitive slide content, and writing clear prose in Word.

Pricing

Grok 4.5 is delivered at an incredibly competitive cost compared to other leading models. Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens. The model also achieves roughly 2x the token efficiency of comparable leading models, solving tasks in under half the number of steps. Overall, Grok 4.5 delivers the highest intelligence per unit of time and cost.

Getting started

Grok 4.5 is available today in Grok Build, in Cursor on all plans, and from the SpaceXAI console. Simply grab an API key and get started in a few lines of code:

Note: Grok 4.5 is not yet available in the EU in any SpaceXAI products or the API console. EU availability is expected in mid-July.

Create an API Key

Start building with Grok 4.5 today via the SpaceXAI API.

API Docs

Read the docs and integrate Grok 4.5 into your stack.

Try it in Grok Build for free

We’re offering free Grok 4.5 usage for a limited time in Grok Build and Cursor. Get started today at x.ai/cli.