LUmBULtERA

I have friends and family who use Gemini, but entirely because their Pixel phones came with a year of it for free. No other reason, and they will most likely never pay for it.

The subsidizing thing is repeated over and over without proof. Personally, I doubt they're actually subsidized.

Maybe, but Tibo said something on Twitter last week making it sound like it might not be.

Given how OpenAI got rid of their 5-hour limits and reset weekly limits so often, is Kimi really undercutting them on effective price?

Sure, but at those API rates it doesn't beat the subscription plans for OpenAI.

Literally no UI changes when I toggle between the two. It's 100% identical, one is not less technical than the other.

Also, when you toggle btween ChatGPT Work and ChatGPT Codex, nothing changes. This is super confusing.

This is the weirdest thing. It's completely unclear even if I was using the app for "work" rather than "coding" why I wouldn't just use the "Codex" option. Why even bother toggling? Can Codex actually not do any of the things in "Work"?

They mean the same thing as GPT-5.6-Max, GPT-5.6-Plus, and GPT-5.6-Fast but require base knowledge for an analogy.

But do they though? When do you use GPT-5.6-Max-Low vs. GPT-5.6-Plus High? Or GPT-5.6-Fast-Xhigh? What's the Pareto optimal choice (outcome and price)? According to the benches it seems to bop around and the even if the benches are accurate the best choice isn't always consistent.

Ping ponging to ChatGPT/Codex for 5.6 launch to test that out... Hopefully its out today or soon.

There could be a whole host of reasons. It may be during launch that compute is re-allocated from training to inference so that all users can try Fable. Soon that compute will re-allocate back to training until they can get more compute.

Huh, super interesting. Thanks!

Did Anthropic have Opus 4.8 and Sonnet 5 switched in the Agentic Search chart at first?

That's yet to be determined. I think a lot of open-weight models are benchmaxxed and their usefulness for many tasks are not represented by those.

Fair to include A&G, but their marketing and training cost is about generating future growth. People say that the subscriptions are highly subsidized as if its fact -- it's certainly not a well-established fact. These people simply don't know the truth one way or another, but portray their not well supported opinion as fact.

The craftsman, who may otherwise be profitable, also has investment costs that cause them to show a loss for some time.

Day 2 the craftsman has not made up for the investment/loss of their equipment. Not a useful example.

OpenAI inference revenue exceeds its cost of inference by a good margin in 2025 (https://cdn.arstechnica.net/wp-content/uploads/2026/06/opena...)

Data from OpenAI shows their 2025 inference revenue exceeded their cost of inference by a good margin (https://cdn.arstechnica.net/wp-content/uploads/2026/06/opena...). Saying this is being subsidized is like saying any investment in future productive assets is "subsidized".

The API inference cost to customers is not the actual cost of providing inference, and the cost of providing API inference need not be the cost of providing subscriber inference.

The API inference cost to customers is not the actual cost of providing inference, and the cost of providing API inference need not be the cost of providing subscriber inference.

My comment is about your statement "serving these tokens without paying for training is already expensive"...

One thing we do know from OpenAI's leaked financial document is that they are already profitable on inference, though that data is not broken down by cost and revenue of API vs. subscription. One important factor is that subscription inference can be optimized in ways to reduce cost (e.g., usage limits, batch optimization around API-prioritized inference, etc...). I think simply we do not know the actual cost of subscription interference for SOTA models.

Deepseek's models are open-weight and hosted all over the world, how would blocking deepseek's web sight do anything to stop its model's use?

Except there are plenty of inference providers worldwide (including the US) that serve open-weight models that are not subsidized, and are reasonable in cost. Or is your claim that those are all running at a loss?

3. We're massively overusing SOTA models. As long as you're on a subsidized subscription, you can use Claude Opus 4.8 high to write blog article meta descriptions. If you paid by token, you wouldn't do that.

This idea that the subscriptions are subsidized is repeated over and over, but I've never seen any proof of this. It seems to be entirely based on the inferred API cost the subscription usage could give you, but there are a lot of assumptions needed for that to follow.

Subscription inference can also be cheaper than the cost of API inference if the provider wants it to -- providers can do flexible scheduling for subscription inference for example, around API inference, to lower its cost and get better utilization of the hardware.

Fair enough, there is not strong specific evidence to the contrary except about overall inference being profitable for OpenAI (as well as the open weight model providers hosted throughout the world).

How is it not easy to get the coding plan?

My impression is that individual subscriptions are the loss leading hook

Except there is no evidence of this at all, just people comparing API and subscription pricing. The leaked financial info for OpenAI shows inference is profitable right now, though it does not show a distinction between subscription and API revenue... but if subscription revenue was so lossy, it would hard for total inference to still be profitable.

I've been testing M3 for agentic tasks on Hermes and it just gets way too confused. I have really poor result from it compared to GPT-5.4 mini/regular or GLM-5.2 (and even 5.1).

Aside from Codex and Claude, what other subscriptions do you think provide the most value?