They have maxed out the base data with the given model architecture and scale "intelligence" through inference-based RL.

Notably, in order to achieve a specific quality, the model has to "reason" for every prompt proportional to the training inference.

This is arguably the case of diminishing returns being reached, as answer token prices saturate the willingness of customers to pay.

They've run into a wall.

If you compare Fable 5.2 Medium to Opus 5.5 high (settings that make sense - max is universally considered rarely worth it), you will see that Fable 5.2 is SUPPOSEDLY very cost competitive: https://artificialanalysis.ai/models/claude-fable-5-2?models...

In case you're unaware, this is for the max setting. The (recommmended) high setting uses only 80k tokens.

Depends highly on the task. It does pretty well on my benchmark[1] in terms of cost (and success rate).

1 - https://bench.killswitch-lang.org/

It beats opus 5.5's max effort with less cost, which is a step forward if you don't want to crank it to Ultra Max Settings.

The amount of compute it cost to get the best previous performance from anthropic went down. Nobody is putting a gun to your head and forcing you to go further down the paretto frontier if the current level of intelligence/task is acceptable to you.

I think we will beat the cost with Hardware. Faster, more efficient hardware. From what I have seen we have only found ways to make it smarter and not more cost effective.

I'm curious to see the next generation of dedicated AI processors and how they perform.

The frontier chart there makes gemini carbon look really weird. Low is worse and more expensive than medium, and high is worse and more expensive than medium... funky curve like that implies a messed up model or weird benchmark.

Well they do have an IPO coming up.

Most people never run out of plan usage, so the cost per task matters relatively little.

It depends on the task though. If your cheap task can't solve it

Why does AA choose ~$3.1 per task to be the breaking point for "most attractive quadrant" highlight? Surely the most attractive section is just being close to the entire frontier?

> it shows how quickly you can go back and forth with a model

Another member of the pro-REPL / anti-agentic tribe. There are dozens of us!

I'm sure that they get early access to run the benchmarks, it wouldn't have time to finish running if they were running it at public release

Someone on a thread was complaining they haven’t really needed the newer models because Opus 4.6 was good enough.

I had to remind them 4.6 only came out February this year. In a normal world 4.6 is a newer model! A model 6 months old is seen as ancient!

>At least Anthropic should not bill me for the refusal (as of my last tests, refusals still comes out of my quota)

There's a setting to not degrade to a lower model on refusals and instead pause the chat.

Opinions are my own.

I think it's an unwarranted assumption that a temporary benchmark lead is the focus. I think using our resources efficiently to provide helpful solutions is more important.

This model is so far off the efficiency frontier that I personally wouldn't want to use it. The wait would reduce my productivity instead of increasing it.

So is Google still in the race? Does anyone actually uses Aragon or Carbon or whatever model they released for coding?

Optics and incentives. Google will turn on the R&D the day after their Anthropic investment is realised, maybe $80-150bn on a 1.2tn valuation, plus >100bn in spending commitments.

This page is showing Max effort, which is much more expensive.

Max should not be used unless you really need every bit of perf, it's deep into diminishing returns territory.

It seems to use way, way more tokens to complete the tasks