There was a lot of Grok related news this week, beginning with Musk saying this week that Grok Bot will send each task to whichever back-end model does it best, including Anthropic’s Claude Opus 5.5.
In addition, Grok Bot can now search, read and monitor X without a connector, and it gained a read-only Plaid connector to serve as your private CFO. In addition, Grok 4.7 managed to top the VulcanBench Frontier v4 Coding Leaderboard, putting it in contention with leading frontier AI.
Grok Bot Gains Task Routing
Musk said on Oct. 7 that Grok Bot will not lean on Grok alone. In a post on X, he said SpaceX will use the best back-end model for each task and named Anthropic’s Claude Opus 5.5, Midjourney and Suno.
Important note regarding Grok @Bot:
Going forward, @SpaceX will use the best back end model for any given task, including Claude Opus 5.5, MidJourney, Suno and other leading APIs.
Whatever is most likely to give you the best outcome.
— Elon Musk (@elonmusk) October 7, 2026Anthropic is also a SpaceXAI compute customer. The two companies signed a deal on May 15 that gives Anthropic access to compute at Colossus 1. SpaceXAI is likely taking advantage of this shared compute to get discounted access to Opus 5.5 for Grok Bot.
Midjourney, on the other hand, competes with Grok Imagine, which makes images and video. SpaceXAI has previously mentioned that Grok’s multimodal features, including image and video generation, are lagging compared to competitors today.
Suno, also noted by Musk, generates music, which Grok has no model for at this time. Asking Grok to make music or sound effects previously would relegate the task to Cursor Cloud, which would pull on a third-party model to generate audio.
Two more posts later that day detailed what the routing schema looks like under the hood.
Grok @Bot will use whatever achieves the best outcome for users.
Simple questions will route to small, fast models. Questions with complex answers will route to large models.
— Elon Musk (@elonmusk) October 7, 2026Simple questions go to small, fast models and complex ones go to large models. Musk said most requests will run on a fast version of Grok 4.8 “when that comes out.” He did not say when that will be.
However, most @Bot requests are pretty simple and will be handled by a lightning-fast version of Grok 4.8 when that comes out.
Operating principle is to give Grok Bot users the best possible combination of speed & intelligence.
— Elon Musk (@elonmusk) October 7, 2026Grok Bot’s help docs already describe this setup. Grok Bot is not tied to one model and picks per task between first-party xAI models and third-party models from providers on Cursor’s sub-processor list. Users get no say. There is no model picker, per user or per request, and a team or user’s Cursor model allowlist or blocklist does not apply to Grok Bot.
Usage screens log everything as Grok Bot usage, with no model name attached, and a request that lands on Claude Opus 5.5 counts the same against a plan’s limit as one Grok handles. Privacy Mode covers every model in the mix, so customer data is not used for training. Zero Data Retention follows Cursor’s existing agreements with each provider, though providers may still run abuse and safety classifiers and store data those classifiers flag, according to the Grok Bot security docs.
Grok Bot Can Now Monitor X, No Connector Needed
Grok Bot’s X access changed the same day. The official @bot account posted a 12-second video captioned “Grok Bot can now search, read, and monitor X.”
Grok Bot can now search, read, and monitor X. pic.twitter.com/KTYcATvrG4
— Grok Bot (@bot) October 7, 2026Grok Bot’s X connector already searched posts, read timelines and checked mentions when it launched in August, but it needed an X sign-in and ran on X API credits. The new version needs no connector and is available to all Grok Bot users. Monitoring is the new part: Grok Bot can track feedback on a product, follow a breaking story or send a weekly roundup of an industry, similar to the Premium+ Radar feature on X, but built into Grok Bot and smarter.
Musk Pitches Grok Bot as a Personal CFO
Musk pitched another job for Grok Bot on Oct. 6. Quoting a demo, he posted: “Grok @Bot can be your personal chief financial officer!”
Grok @Bot can be your personal chief financial officer!
— Elon Musk (@elonmusk) October 6, 2026The product behind it launched Sep. 26, when the @bot account added a Finance connector to the Grok Bot marketplace. Bank, card and investment accounts link through Plaid. Access is read-only, Grok Bot never sees logins, and users can unlink at any time. Grok Bot cannot spend any money through the Plaid connector either. The connector launched in the U.S., and Plaid’s availability is limited to the U.S. only at this time, so international users will need to use their bank’s MCP server or available data files for a more limited functionality.
Teslaconomics noted that they have linked about 30 accounts at roughly eight institutions, 13 of them at Chase, and asks Grok Bot for cash across accounts, card due dates and its investment mix.
Musk made a bigger promise in late August, before the connector launched, telling a user, “If Grok Bot messes up, we will make you whole.” SpaceXAI’s Terms of Service contain no such guarantee, and the company has not said whether write or payment access is coming.
Grok 4.7 Tops VulcanBench Frontier v4 in Cursor
Grok 4.7, released Sep. 21, now holds the top three columns on VulcanBench Frontier v4, an open coding benchmark where agents rebuild a retired program whose behaviour has drifted from its spec across 23 tasks. The board last updated on Oct. 4.
Run through Cursor‘s agent CLI from Oct. 1 to 3, Grok 4.7 scored 93.15 at Extra-high, 92.71 at High and 92.30 at Medium, according to the effort report. Low scored 89.42, 12th on the board. The best non-Grok column is Claude Fable 5.1 Max at 91.84.
Grok 4.7 had different judges from the rest of the board. Other columns are graded for code quality by Muse Spark 1.3 and Grok 4.6. Grok 4.7 was graded by Muse Spark 1.3 and OpenAI’s GPT-6.1 Sol, because Grok 4.6 is not a neutral judge for an xAI model. GPT-6.1 Sol rates Grok 4.7’s code 5.4 to 6.3 points higher than Muse does, so its scores are not strictly comparable with other columns.
Scored by Muse alone, Grok 4.7 still ranks first at Medium, High and Extra-high. VulcanBench lists Grok 4.7’s cost as unavailable, because the runs used a Cursor subscription. On the standard API, Grok 4.7 costs $2 per million input tokens and $6 per million output tokens.
