Independent Assessment of AI R&D Acceleration at Anthropic in October 2026

> The acceleration in late 2025 was likely influenced by non-AI factors. We investigated a variety of factors which could have contributed to the observed acceleration in late 2025, but details of many of these factors have been redacted.

Of note, "the acceleration in late 2025" believe has to do with Opus 4.5 which in my view and from Metr's analysis[0], was the point where their tests stopped working because the model was so good that doing work without using it took much more time.

> We guess that a majority (perhaps ≈80%) of the current (October 2026) acceleration is due to AI.... We believe AI’s contribution has grown relatively more important over time, and do not think the other factors we investigated would explain the sustained acceleration. The evidence of speedup from the usage of AI we describe below also seems compatible with a large contribution.

In a mini bet, I'd say that Anthropic is going to win by a large margin in what's going on. I've used the other models and no matter what the benchmarks are, places like OpenAI, Gemini, or the Deepseek / Kimi models haven't come close to the ability of the Claude models and harness. If what they say in this post is right, with "We did not identify any convincing bottlenecks to continued rapid progress.", this could be a sort of game over for the other models.

Even with Opus 5.5 in the past month and a half has been a changer in its abilities. In my dev process, I can give it much more freedom and trust with the workflows and connections I have to do the valid work to where I can do more, bigger tasks at once. I can't tell you how curious I am with the Opus 5.6 and whatever Model 4 is going to be called (along with the cost!) will do for dev work. And if this keeps increasing, and if there are subtle tricks that Anthropic has that the other companies don't, what that means for finding solutions and overall value and if the others are ever going to be able to catch up in any way.

All speculation, but what a time to be a dev in these times and get the first hand experience of all of this!

[0] https://metr.org/blog/2026-02-24-uplift-update/#wider-adopti...

Although unsurprising given that there’s perhaps a trillion dollars at stake, it is disappointing that this article opens with “Anthropic’s redacted the best parts”. The table of shared/redacted sources is unfortunately dominated by no/partly.

> As of October 2026, Anthropic is likely improving the capabilities of their models on well-specified tasks at around twice the rate they did between March 2024 and October 2025. This estimate is given with high uncertainty; we would not be surprised if the true acceleration was closer to 1.3x or 3x the historical rate. > A majority of this acceleration today is likely attributable to Anthropic’s extensive use of AI to do R&D, though there is significant uncertainty and other causes likely played a large role in acceleration in late 2025. > The lack of observed bottlenecks or strong diminishing returns means there is a significant risk that further capabilities progress could lead to even faster acceleration, though, as we say below, we may be missing important bottlenecks.

Cool to hear it's over for student before it ever started

The most damning report indicating the severe trajectory towards RSI and ASI at frontier labs, the soon to be most impactful event in the world, and it’s barely a whisper on HN. Typical

I think that this report is an empty shell with the purpose to have Anthropic look good with a lot of "growth" potential before the possible IPO look good without saying anything concrete and without making promises that could later be considered "deceptive statements".

I like that the authors appears to be honest and well meaning by ensuring to make it clear every where that they basically know nothing or can't disclose it.

A data point they use and that shows a nice growing chart is: "Informal self-reported uplift over time from Anthropic employees"

What a joke to pretend it is anything close to a valid scientific data point.

It is "self reported", a "uplift" feeling without concrete measure, and from "Anthropic employees" that are the one creating the tools that pretend to be evaluated here.

You see someone saying "my work is so much horse shit that i would not use it myself", "or you pay me to improve that, but it doesn't make me more efficient or worse". And also, even if you wanted to be honest, how would it look when you have to face the top management after. Not considering that their "stocks" value might depend on their reply.

And you have the following that is so totally bullshit crazy:

   "We did not see evidence for significant bottlenecks which would prevent further capabilities progress due to limits on algorithmic progress, training compute, experiment compute, or data in the next 12 months."

You don't know what you don't know, but we think that there is no barrier for the capability being improved when currently we are at the maximum we could have reached at this point.

There is a ram shortage worldwide, it becomes hard to find supply in GPU or build data centers, but they don't see training compute limitation in the coming year...