← Back to Pulse PULSE. brief
Gemini 4 and Muse Wonder Are HUGE… But Claude DESTROYS Them

Gemini 4 and Muse Wonder Are HUGE… But Claude DESTROYS Them

Riley Brown26 min2026-10-16 ▶ Watch on YouTube
ℹ️
Partly verified — 5 of 62 specificsA few specific details here couldn't be independently confirmed against the video. The overall summary is sound, but double-check exact numbers or names before you rely on them.
What this video is
⚡ a 26-minute video, readable in 60 seconds

Meta released Muse Wonder and Muse Wonder Full this week, with Mark Zuckerberg announcing it on X as a frontier model family, and Google followed a day later by rolling out Gemini 4 Argon globally in the Gemini API; Meta's own self-reported benchmarks place Muse Wonder Full second only to Claude Opus 5.5 on the Personal Superintelligence Index. The video also reports a wave of other updates the same week, including GPT 6.1 Sol, Grok updates, Claude Code and Codex changes, and new GrokBot features. The creator argues these releases show Google and Meta catching up to OpenAI and Anthropic on the frontier, but in their own hands-on testing they found Gemini 4 Argon's and Muse's game-building results worse than ones built with Claude Opus, leading them to conclude Opus 5.5 is still the best model right now.

Confirmed: Google released Gemini 4 Argon and Meta released Muse Wonder and Muse Wonder Full this week (timestamp not stated).
Key takeaways
+ 54 more takeaways
  • Confirmed: [01:57] the Muse agent, named Rick, is a free app positioned as a competitor to Grokbot and GPT Dots.
  • Confirmed: [02:25] Muse Wonder is priced at $4 per million input tokens and $20 per million output tokens, the same price as Claude Opus 5.5.
  • Confirmed: [02:54] Muse Code is used in the terminal by typing "Muse," similar to invoking Claude in the terminal.
  • Confirmed: [03:01] the model used was Muse Wonder 1.0, priced at $4 per million tokens.
  • Confirmed: [03:26] setting up Muse Code requires signing in with a Meta account during terminal setup.
  • Confirmed: [03:47] the day after Meta's release, Google rolled out Gemini 4 Argon globally in the Gemini API on Tuesday.
  • Confirmed: [04:08] Google's Ultra Sub costs $100 per month and is required to access Gemini 4 Argon inside Antigravity.
  • Confirmed: [04:33] the speaker built a Call of Duty style game in Antigravity using about 10 prompts.
  • Confirmed: [06:25] Gemini 4 Argon launched at an intro price of $2 per million input tokens and $10 per million output tokens.
  • Confirmed: [06:43] after November 30th, Gemini 4 Argon's price rises so all compared models converge at $4 per million input tokens and $20 per million output tokens.
  • Confirmed: [09:02] "Update 10" is a new redesigned Claude desktop app.
  • Unverified: [09:58] "Update 9" is a rumor that Fable 5.5 will release either the next day, the day of filming, or the following week.
  • Source cited: [10:18] the speaker cites Andrew Curran, described as a source that "generally turns out to be right," as the basis for thinking Fable 5.5 might arrive the same day as filming (Thursday).
  • Source cited: [11:39] per Tibo, who runs the Codex/ChatGPT desktop app product, OpenAI is issuing a new feature or reset every day for 28 days since Dev Day, currently around day 12.
  • Confirmed: [12:49] GPT 5.1's context window was updated from 272k to 500k.
  • Confirmed: [12:49] the ChatGPT desktop app now uses about 60% less disk space.
  • Confirmed: [14:55] for the past two weeks, the speaker's morning routine has been walking for an hour while talking to their "dot."
  • Confirmed: [15:05] the dot has access to the speaker's computer running locally and can still work in the cloud if the computer is off.
  • Confirmed: [15:13] DOTs allows spinning up multiple Codex sessions in sequence from a single conversation thread.
  • Confirmed: [16:09] "Update 15" lets users create their own email inside GrokBot by asking it to set one up.
  • Confirmed: [16:34] a GrokBot email can send emails to team members, sign up for newsletters, and sign up for services like Gmail on the user's behalf.
  • Confirmed: [16:53] GrokBot now connects natively to X, letting users search Twitter without an X API key, extra payment, or connecting their X account.
  • Confirmed: [17:24] the speaker set up a Claude Code webhook skill letting Claude Code send prompts to GrokBot to search X, such as compiling a table of OpenAI's resets and updates.
  • Confirmed: [18:02] X now lets users search Twitter/X natively via GrokBot, which the source compares to how Muse gives native permissions for Instagram and Facebook.
  • Confirmed: [18:27] Agent Native's resources page has a guide for connecting Claude Code to GrokBot via webhook, usable from any agent including Codex.
  • Unverified: [18:30] the source says the Claude Code to GrokBot webhook setup can reportedly be completed in under two minutes.
  • Confirmed: [18:48] a new voice model, Grok Voice 3.0, was announced and already released in the API.
  • Source cited: [18:56] an X post from @SpaceXSI announced Grok Voice 3.0 with a benchmark table against Grok Voice 2.0, GPT-Live-1, and Gemini 3.8 Live.
  • Unverified: [19:04] Grok Voice 3.0 reportedly ranks #1 on Artificial Analysis's Speech-to-Speech task and scores 97% on Full Duplex Bench, available free via a link at x.ai/news/grok-voice-3.
  • Confirmed: [20:26] GrokBot's cloud computer for agent use is now faster and higher resolution.
  • Confirmed: [20:26] email drafts in GrokBot now have correct formatting with easy CC editing.
  • Confirmed: [21:28] Max Blade spent three weeks and two Claude Max accounts vibe coding a 20-person battle royale game called Kong Royale.
  • Confirmed: [21:28] Max Blade live-streamed chat-driven changes to the game in what he calls "hive coding," including 60 changes made overnight after Opus 5.5 was used.
  • Source cited: [22:05] Max Blade said this was when "hive coding" was first invented, where he live streams a vibe-coded video game and chat suggests changes.
  • Confirmed: [22:16] the game Max Blade live streamed is a Fortnite-style battle royale game.
  • Confirmed: [22:32] Max Blade takes the best suggestions from chat, lets the game update overnight, then plays the updated version.
  • Confirmed: [22:56] Max Blade is premiering a "non-jank" battle royale on Thursday, October 15th at 1pm Eastern.
  • Confirmed: [24:02] Max Blade created a multiplayer Call of Duty style game that he vibe coded and played live with 19 other people, adding new maps and guns every three prompts.
  • Confirmed: [24:40] Xbox's CEO announced "Xbox Labs," a group exploring using AI to let fans recreate and remix Xbox titles like Halo.
  • Take (speaker): [24:40] the speaker likened Xbox Labs to Halo's Forge mode "but with AI."
  • Confirmed: [25:59] the agentnative.inc updates page lets viewers see every item mentioned in the video along with links, organized by week.
  • Confirmed: [26:01] agentnative.inc's "Agent Landscape" page lists 65 AI agent tools across 17 categories.
  • Take (speaker): the releases of Gemini 4 Argon and Muse Wonder prompted claims, including from the creator, that Google and Meta have caught up to OpenAI and Anthropic on the frontier (timestamp not stated).
  • Take (speaker): [04:33] the speaker judged the Antigravity-built Call of Duty style game worse than one he built with Opus 5.1, describing its enemies as "terrible."
  • Take (speaker): [05:20] the speaker said the game is "not quite as good as Opus and Astra."
  • Take (speaker): [05:24] the speaker called the Gemini 4 Argon built Call of Duty style game "much worse," in his opinion.
  • Take (speaker): [05:42] the speaker said using Opus inside Cursor burns through credits quickly.
  • Take (speaker): [05:47] the speaker said Opus 5.5 inside Claude is hard to hit the usage limits on the $100 to $200 per month plan.
  • Take (speaker): [07:01] the speaker said Opus 5.5 is the best model in the world right now.
  • Take (speaker): [08:36] the speaker said GPT 6.1 Sol is better than Gemini 4 Argon at almost everything, and that the Codex app is way better than Antigravity.
  • Take (speaker): [08:50] the speaker said Opus 5.5 versus Fable 5.5 pricing shows Fable is not much cheaper and not better at anything.
  • Take (speaker): [09:47] the speaker called Claude Desktop running Claude Opus 5.5 "the dream stack right now."
  • Take (speaker): [14:07] the speaker said doing tasks hands-free with AirPods while talking to a coordinator agent running tasks in parallel is "extremely productive."
  • Prediction: [10:43] the speaker predicts Fable 5.5 will likely be priced the same as Fable 5.1, about 2.5 times more expensive than Opus 5.5.
Confirmed: [17:24] the speaker set up a Claude Code webhook skill letting Claude Code send prompts to GrokBot to search
Confirmed: [17:24] the speaker set up a Claude Code webhook skill letting Claude Code send prompts to GrokBot to search ▶ 17:38
Confirmed: [18:27] Agent Native's resources page has a guide for connecting Claude Code to GrokBot via webhook, usable f
Confirmed: [18:27] Agent Native's resources page has a guide for connecting Claude Code to GrokBot via webhook, usable f ▶ 18:29
Source cited: [18:56] an X post from @SpaceXSI announced Grok Voice 3.0 with a benchmark table against Grok Voice 2.0, G
Source cited: [18:56] an X post from @SpaceXSI announced Grok Voice 3.0 with a benchmark table against Grok Voice 2.0, G ▶ 18:56
How this brief was shaped: News Analysis / Commentary · confidence High

Transcript is a weekly AI news roundup naming specific releases (Muse Wonder, Muse Wonder Full) with the creator giving direct opinion and advice ('use Opus 5.5, it is the best model'), and OCR shows an official product announcement plus a benchmark comparison table the creator is reacting to.

The lens sets this brief's structure, never its facts — every claim is held to the same citation and fact-check standard.

Jump to a moment
0:00Intro0:36Meta Muse Wonder3:11How to Use Muse Wonder3:47Google Gemini 4 Argon4:18Argon Call of Duty Test5:32Where to Use Argon6:14My Thoughts on Argon & Muse7:51Don’t Feel Behind8:53New Claude Desktop Design9:57Fable 5.5 Rumors11:39Codex Resets & Updates13:29New Dot Voices14:02Working with AirPods16:03GrokBot Email16:53GrokBot X Search18:47Groq Voice 3.020:24Faster GrokBot Computer21:21Vibe-Coded Video Games24:39Xbox Labs
Their links, sorted & clickable
🔗 Other links2All of this week’s updates and linksagentnative.incFree guides and the Agent Landscapeagentnative.inc
← Back to Pulse Dashboard
Was this brief useful?