I'm a Pro subscriber since Bard first came out.

I did the obligatory "SVG Pelican" test so you don't have to - with Argon. You can check my post history and see the amazing results I got with 3.8 Flash not too long ago btw!

Anyway, I'm extremely disappointed by the results, and I know this isn't a great benchmark, but it just did so poorly... I tested it on High, Medium, and Low effort.

The High effort (only one prompt) took 19% of my 5 hour quota!! And the result was horrible (much worse than 3.8 flash). All 3 tests (high, medium, low) consumed 51% of my 5 hour limit. Just 3 prompts.. and all 3 results were objectively horrible.

For reference, 3.8 flash on High with the same prompt consumed only 1% of my usage! And got much better results.... How does this make sense?

Picture 1: Argon High Result

Picture 2: Argon High usage after first test (19% consumed)

Picture 3: Argon Low result

Picture 4: Argon Low usage after 2nd test (another 15% consumed)

Picture 5: Argon Medium Result

Picture 6: Argon Medium Usage (another 17% consumed)

Picture 7: 3.8 flash High Result

Picture 8: 3.8 Flash usage (only 1% used!!)

This makes no sense to me, and the usage limits are completely broken on Argon.

Yeah I know it's a silly test, but still incredibly disappointing for multiple reasons (results and insanely nonsensical limits usage).

I did one final test on 3.1 Pro High (results also bad) and it used 9% of my 5hour limit. I'm not adding those screenshots to the post but that's also an insane amount of usage for the 3.1 pro model which I used to be able to use without any worries for hours on tasks.

Disappointed.. especially with the crazy limits.

Homoncrus from full metal alchemist

What should I ask him?

1% before. 5% after. I am in IT so I utilized the following prompt:

Role and Purpose

You are an expert wireless network architect and senior telecommunications engineer specializing in IEEE 802.11be (Wi-Fi 7) specifications. Your task is to provide an exhaustive, technically rigorous, and deeply detailed explanation of the Wi-Fi 7 "time to wait" feature, formally known as Restricted Target Wake Time (R-TWT), along with related scheduling and medium access control mechanisms.

Core Objectives

1. Explain the operational mechanics of Restricted Target Wake Time (R-TWT) in Wi-Fi 7 and how it improves upon the legacy Target Wake Time (TWT) found in Wi-Fi 6 (802.11ax).

2. Detail how access points (APs) and stations (STAs) negotiate wake intervals, service periods, and resource allocation to minimize medium contention and power consumption.

3. Address the deterministic scheduling aspects, how it handles latency-sensitive traffic, and its role in crowded environments (e.g., enterprise deployments, IoT dense grids).

4. Outline the technical parameters, frame structures, and operational states associated with R-TWT.

Required Depth and Structure

Your explanation must not be a high-level summary. Cover the following dimensions thoroughly:

- Evolution and Context: Explain why traditional contention-based channel access (EDCA) falls short in dense deployments and how TWT evolved into Restricted TWT (R-TWT) for IEEE 802.11be.

- Mechanism Mechanics: Step-by-step breakdown of how R-TWT works, including parameter set negotiation, allocation types (sub-tier vs. non-sub-tier), and how devices synchronize sleep and wake states.

- Determinism and Latency Reduction: Explain how R-TWT provides predictable latency, jitter reduction, and resource reservation for time-critical applications (such as AR/VR, industrial automation, and real-time streaming).

- Interactions with Other Wi-Fi 7 Features: Discuss how R-TWT coordinates alongside Multi-Link Operation (MLO) and Multi-Resource Unit (Multi-RU) puncturing.

- Edge Cases and Limitations: Address behavior during high interference, device fallback scenarios, and handling of unacknowledged frames or missed wake windows.

Constraints and Guidelines

- Maintain absolute technical accuracy rooted in the IEEE 802.11be standard draft and finalized specifications.

- Avoid surface-level generalizations, marketing fluff, or vague analogies. Use proper engineering terminology (e.g., MAC layer, Service Period, Trigger Frames, SP Duration, Wake Interval).

- If certain implementation specifics vary across commercial silicon vendors, explicitly note these variances.

- Do not truncate your explanation; cover all technical nuances comprehensively.

-----

The response was pretty long and detailed!!

Regardless of reasoning mode the current cap usage of the 5 hour window is 5% per question, use that as you will. A per question max use cap has been in place since the May 29, 2026 X post from Josh Woodward mentioning they set a cap per question.

Meaning you get at the very least 20 questions per 5 hours. Expect roughly 30 questions per 5 hour window with a range of 20-60 depending on complexity.

Currently you get 20 full five-hour windows per week (full usage of 5 hour quota only takes 5% weekly), so that gives you nearly 3 full windows a day and 400 questions weekly minimum if you are able to fully utilize all the 5 hour windows. Essentially the weekly limit is not a problem.

For mundane extremely simple queries, regardless of reasoning effort (tested with a simple hello) the first question in the conversation seems to take 4-5% usage. The second question seems to take 2-4% and subsequent questions take 1% or even 0% (rounded) quota per question.

Let me know if you find anything different. This may change as well.

Google's rate limits are a complete joke compared to OpenAI right now, let alone Claude. It is crazy because it was the exact opposite a year ago, though that is a topic for another day.

Being slow, falling behind over the past six months, and lacking a true flagship model has made their Ultra tier one of the worst value subscriptions out there. I bet almost nobody is actually paying for it. Now that they finally managed to cook up a flagship model with decent benchmarks, they went with the most short-sighted marketing play possible. They locked Pro and Plus users out and made it an Ultra exclusive. I guess Pro users should thank Google for showing mercy by letting us use it inside their terrible Gemini app, while Plus users get nothing at all. The real cost here is alienating the vast majority of Pro and Plus subscribers, along with potential paying users.

Honestly, Google seems to have zero long-term vision with their go-to-market strategy. Desperate to steal market share from ChatGPT, they started by handing out way more freebies than they could actually afford. First it was the insanely generous free API tier that got nuked earlier this year. Then came the 6 to 12 months of free Pro bundled with Pixel phones, 12 months free for students, and the market getting flooded with 18-month Pro promos from Jio. These promos always last half a year to a year. By the time Google realized their compute and R&D budgets could not handle the load, those free accounts were already everywhere with more than half their time left.

So what does Google do? Pull the plug on the Jio partnership? They do not have the guts. Stop the student and Pixel freebies? Then they could not brag to executives about skyrocketing subscriber numbers. Instead, they went with a blanket nerf to rate limits across all paid tiers, leading up to how they are shafting Pro and Plus users today. The only reason they did not make a new "Max" tier is because they never handed out Ultra for free in bulk. If they had, I guarantee they would invent a new tier just to gatekeep the good stuff. This only makes people do the math, realize paying tens or hundreds of dollars for pathetic limits is a rip-off, and drives them straight to the gray market. You can blame the gray market deal hunters all you want, but Google created this mess. I refuse to believe nobody smart at Google saw this coming, especially after the Jio deal. Chasing short-term metrics for the suits pushed them into making these short-sighted decisions.

At the end of the day, subbing to Google Ultra just for Argon is really not worth it. Do not fall for the FOMO and impulse buy, especially since I hear the rate limits on Ultra are still tiny and nowhere near what Claude gives you for Opus 5.5 on their normal plans. Maybe after the Jio promo ends later this month and they stop handing out Pro like candy, Google will finally bring rate limits back to a usable level. Until then, there is no way I am paying full price for Google.

It's already an hour or two away from being officially released. There's no word yet on what model or level it will be. We'll wait and see. About an hour left.

I just checked the ios app and I’m getting some new features like the chat and task options in the same area as opposed to having to go to Spark.
I’m also getting the Argon option even though I don’t have any of the normal model options (I presume that the Argon model is the equivalent option of Pro so Gemini 4 flash isn’t there).

If someone has some good suggestions for using the Argon model for something that’ll be worthy for its use, please let me know!