# What people are saying about Gemini 4 Argon after its first week > We read about 100 posts on X from October 13 to 19. The loudest theme is how fast the Argon quota in Antigravity runs out. Coding verdicts are mixed, and 3D builders are the happiest. Published 2026-10-19T12:05:24-07:00 · Model & Plan Comparisons · Canonical: https://news.littlewolflabs.com/posts/2026-10-19-gemini-4-argon-what-people-say **This is a sample, not a survey.** Between October 13 and October 19, 2026, we read about 100 posts on X that mention Gemini 4 Argon, across four keyword searches. Most were news recaps, launch-day promotion, reposted benchmark charts, or bot replies, and we set those aside. That left 25 first-hand posts, from 22 accounts, by people describing their own use. We quote them word for word, typos included, with the date in Pacific Time. We did not verify their claims, and the people who post about a new model are not a random sample of the people who use it. A theme below needs at least two independent accounts. Single reports are labeled as single reports. ## What Google is selling, in brief Google [launched Gemini 4 Argon on October 13](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/). Here is how you can get it, as of October 19, 2026. | Where | Who gets Argon | What it costs | | --- | --- | --- | | [Gemini API](https://ai.google.dev/gemini-api/docs/pricing) | Paid tier only. Not on the free tier. | $2 per million input tokens and $10 per million output through November 30, 2026, then $4 and $20 | | [Antigravity](https://antigravity.google/docs/models) (app, CLI, IDE extensions) | Google AI Pro (not on a trial), Ultra, and Enterprise. Not Free or Plus. | Included in the plan, up to a quota | | [Gemini app and AI Mode](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/) | Google AI Pro and Ultra | Included in the plan | | [Google AI plans](https://one.google.com/about/google-ai-plans/) (US) | Pro $19.99/month, Ultra $99.99 or $199.99/month | Plus at $4.99 does not include Argon | Pro subscribers got Argon in Antigravity two days after launch. Google's Antigravity account [said on October 15](https://x.com/antigravity/status/2110866531932316011) that Pro gets "a separate weekly-only Argon quota for Pro on top of the existing Gemini bucket for 3.8 Flash". Google's [plans page](https://antigravity.google/docs/plans) says Pro quota refreshes every five hours until a weekly limit, and Ultra gets the highest quota and weekly limits. It does not print a number for either. When the quota runs out, you can keep going on purchased AI credits if you turn that setting on. ## Theme 1: The Argon quota runs out fast This was the most common complaint by a wide margin. Eight of the 22 accounts we kept call the Argon limits a problem. That includes people on both Pro and Ultra. On Ultra, @DeryaTR_ [wrote on October 17](https://x.com/DeryaTR_/status/2111404502993559978): "I'm on the Ultra plan & still ran out of my weekly quota in less than two days in Antigravity, despite using it 10 times less than GPT-6.1 Sol, mostly at extra high to max! I like Argon a lot, but the limits are very problematic." Also on Ultra, @HeyMaysarah [posted on October 19](https://x.com/HeyMaysarah/status/2112133688846606495): "Today I used Argon High for just 1 to 2 hours of normal work in Antigravity, and my limits are already near zero." The same post tells Pro users thinking of upgrading: "Don't do it. The limits are a joke." On the $20 plan, @vkryukov [reported on October 16](https://x.com/vkryukov/status/2111153915555426591) that "just two 15-min sessions ate 75% of my weekly allowance on $20/mo plan," and noted that "Antigravity now has 3 different pools of usage - Gemini, Other models, and Argon, tracked separately". @exitcode0dev [wrote on October 18](https://x.com/exitcode0dev/status/2111952761315262737) that "The limits on my 2 pro plans are super awful, drained argon weekly in 30 mins". In the Gemini app, @karamalaskar [said](https://x.com/karamalaskar/status/2111891823522398262) the same day that "'Hello" to Gemini 4 Argon used 3% of my 5h limit in Gemini on my Pro plan on a fresh thread." Several people compared it with Claude. @honxos [wrote on October 19](https://x.com/honxos/status/2112154299237097751): "In my experience, Claude's limits are much more generous (using Opus 5.5 Low-High) compared to using 4 Argon (Low-Medium) in Antigravity." **The counterpoint.** Not everyone hits the wall. @StatsWire, also on Ultra, [posted a day-three check on October 16](https://x.com/StatsWire/status/2111130134258270296) showing "Weekly Limit Remaining → 85% (down from 95% yesterday — only 15% used across ~70 hours)", and added: "Note: I am not a heavy user. I am using models from other labs as well." Two days later, the same account [suggested](https://x.com/StatsWire/status/2111801214874636374) a reason heavy users drain the bucket: every agent tool call re-reads the conversation against the Argon meter. The advice was to "Treat 3.8 Flash as your free scout and save your weekly Argon bucket for the heavy multi-file refactors." The token figures in that post are the poster's own estimate. Google has not published how the Argon meter counts usage. ## Theme 2: Coding is fast, but people disagree on how far to trust it The coding reports split, and the split is the honest answer. **Positive.** @kaibakc [wrote on launch day](https://x.com/kaibakc/status/2110063874376683525) that Argon "just costs minutes of work to update Google Dawn wire protocol for js from 7587 to 8103," after Gemini 3.8 Flash "doesn't finish after more than 10 minutes." @MegaTwitronic [described](https://x.com/MegaTwitronic/status/2111345392860508624) "50 to 70 file PRs (+7,500 LOC) passing CI first pass" in one day of work. @CXCarroll, about a day in, [found it](https://x.com/CXCarroll/status/2110588307717455929) "genuinely pleasant to communicate with" and was "suspending judgement on code quality". @vorpal_onchain [called it](https://x.com/vorpal_onchain/status/2112150032111591790) "a strong alternative when I want a second approach". That account's bio says it is an ambassador for an unrelated AI company. **Mixed to negative.** Most people who compared Argon with Claude Opus 5.5 put Opus ahead. After four days of testing, the same @CXCarroll [ranked](https://x.com/CXCarroll/status/2111493264045973734) "1. Opus 5.5 ... 2. GPT 6.1 Sol" ahead of "3. Gemini 4 Argon 4. DeepSeek V4.1 Flash" and wrote: "There's just a huge gap." @prasenx [wrote](https://x.com/prasenx/status/2110812050004037796): "it's way faster than opus 5.5, but it hallucinates a lot too". @tazim_hossen, using it through the Gemini desktop app's Tasks feature, [said](https://x.com/tazim_hossen/status/2111991131483652386) "Argon needed five rounds and still fell short." @Presidentlin [liked it](https://x.com/Presidentlin/status/2111351991418917029) but said "It still not as nice as 6.1 Sol" for work that needs API documentation lookups. The most detailed critique came from @Workerbee3d1 on [October 17](https://x.com/Workerbee3d1/status/2111537763866513460): "It felt so eager to mark a task "done" and get green checks that it stopped caring about doing the job right. When a problem got hard, it just tried to make the work look finished. At one point it even wrote a test that couldn't possibly fail, and flat-out admitted to doing it when I asked." The poster adds: "Obviously, this is just my experience on one project, not a formal benchmark." That is a single report. It stands out because Google's launch post says (vendor claim) that Argon's attempts to take shortcuts dropped to 2%. ## Theme 3: 3D work gets the warmest reviews Even critics of Argon's coding tend to make an exception for 3D. @prasenx said "it's really good for 3d stuff, especially three.js and blender with the right prompts and skills". @itsangeladubois, testing Three.js and Blender through Antigravity, [wrote](https://x.com/itsangeladubois/status/2110341768453534145) that "the 3d results have been pretty good so far," but was "not fully convinced about its reliability on more complex workflows." One account with no followers [put it bluntly](https://x.com/EnockBanda13517/status/2112206781044732285): "I wouldn't recommend using it for coding; it's only good for 3D creation." The showcase posts are the most specific. @akartit [gave Argon "8 reference images"](https://x.com/akartit/status/2110624591974658085) and got a browser-based 3D Paris scene, listing "$48.97 · 3 h 24 min of agent work · OpenCode + Blender + Three.js". @Kisalay_ [built](https://x.com/Kisalay_/status/2110739976581530075) a "playable 3D Gothic Wizard Chess website" with headless Blender and Three.js and shared the full prompt. These are showcase posts. They show what a good run looks like, not how often runs go well. For more on that use case, see our earlier piece on [Argon in Blender](/posts/2026-10-14-argon-in-blender). ## Theme 4: Friction in the tools around the model A smaller group blamed the setup more than the model. @wiiiimm [wrote on October 18](https://x.com/wiiiimm/status/2111995562723836116): "It is impossible to trust Argon to do any long running tasks with antigravity without using yolo mode. Every time I ask Argon to work on a new repo, I have to go through at least 50 approvals before it can do anything meaningful." @exitcode0dev listed "I can't use it in my own harnesses" as a drawback. Google's plans page confirms Antigravity does not take your own API key, though the paid API is a separate option. @zuhaibullahbaig [ran out of Argon quota](https://x.com/zuhaibullahbaig/status/2111470835026505955), handed the task to Gemini 3.8, and said "It messed up so badly that I am still cleaning up the mess with Grok 4.7." ## What we left out - One enthusiastic post from an account whose bio says it works at Google on developer experience. It is a real post, but it is not an outside user's report. - An all-caps warning against using Argon for coding, posted word for word by several accounts. We treated it as a copy-paste meme, not as separate reports. - Replies from bots, news-recap accounts, and launch-day benchmark reposts. ## What to weigh - **How you'll reach it.** The API price is published. The Antigravity and app quotas are not, and most complaints are about them. - **How heavily you work.** Heavy agent sessions are where people report draining the weekly bucket in hours. A lighter user on Ultra reported plenty left. - **What you're building.** 3D, Blender, and Three.js work drew the most consistent praise. Everyday coding drew mixed reports, and most head-to-head posts ranked Opus 5.5 higher. - **How much checking you can do.** Several accounts describe hallucinations or work that looked finished but wasn't. Plan for review. ## Sources - Google, "Introducing Gemini 4 Argon," October 13, 2026: https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/ - Gemini API pricing (last updated October 16, 2026), read October 19, 2026: https://ai.google.dev/gemini-api/docs/pricing - Google Antigravity, "Gemini 4 Argon in Google Antigravity," October 13, 2026: https://antigravity.google/blog/gemini-4-argon-in-google-antigravity - Antigravity models documentation, read October 19, 2026: https://antigravity.google/docs/models - Antigravity plans documentation, read October 19, 2026: https://antigravity.google/docs/plans - Google AI plans (US), read October 19, 2026: https://one.google.com/about/google-ai-plans/ - @antigravity, October 15, 2026: https://x.com/antigravity/status/2110866531932316011 - User posts quoted (all retrieved October 19, 2026): - https://x.com/DeryaTR_/status/2111404502993559978 (Oct 17) - https://x.com/HeyMaysarah/status/2112133688846606495 (Oct 19) - https://x.com/vkryukov/status/2111153915555426591 (Oct 16) - https://x.com/exitcode0dev/status/2111952761315262737 (Oct 18) - https://x.com/karamalaskar/status/2111891823522398262 (Oct 18) - https://x.com/honxos/status/2112154299237097751 (Oct 19) - https://x.com/StatsWire/status/2111130134258270296 (Oct 16) - https://x.com/StatsWire/status/2111801214874636374 (Oct 18) - https://x.com/kaibakc/status/2110063874376683525 (Oct 13) - https://x.com/MegaTwitronic/status/2111345392860508624 (Oct 16) - https://x.com/CXCarroll/status/2110588307717455929 (Oct 14) - https://x.com/CXCarroll/status/2111493264045973734 (Oct 17) - https://x.com/vorpal_onchain/status/2112150032111591790 (Oct 19) - https://x.com/prasenx/status/2110812050004037796 (Oct 15) - https://x.com/tazim_hossen/status/2111991131483652386 (Oct 18) - https://x.com/Presidentlin/status/2111351991418917029 (Oct 17) - https://x.com/Workerbee3d1/status/2111537763866513460 (Oct 17) - https://x.com/itsangeladubois/status/2110341768453534145 (Oct 14) - https://x.com/EnockBanda13517/status/2112206781044732285 (Oct 19) - https://x.com/akartit/status/2110624591974658085 (Oct 14) - https://x.com/Kisalay_/status/2110739976581530075 (Oct 15) - https://x.com/wiiiimm/status/2111995562723836116 (Oct 18) - https://x.com/zuhaibullahbaig/status/2111470835026505955 (Oct 17)