The Memo - Special edition - Claude Opus 4.6 - Feb/2026

The best frontier model in the world (for Feb/2026)...

Feb 05, 2026

To:      US Govt, major govts, Microsoft, Apple, NVIDIA, Alphabet, Amazon, Meta, Tesla, Citi, Tencent, IBM, & 10,000+ more recipients…
From:    Dr Alan D. Thompson <LifeArchitect.ai>
Sent:    5/Feb/2026
Subject: The Memo - AI that matters, as it happens, in plain English
AGI:     97%
ASI:     0/50 (no expected movement until post-AGI)

Anthropic releases Claude Opus 4.6

Once again, we have this out to The Memo readers within just a few hours of model release. You can rewatch the livestream.

I don’t want to alarm anyone, but this frontier model is far, far beyond human. My testing (including the ALPrompts later in this edition) showed unexpected and complete patterns of responses. Perhaps for the first time, this model feels both superhuman and complete. I couldn’t trick it, it’s incredibly grounded (it knows when it’s being tested on hallucinations), and the overall quality has jumped significantly. One Anthropic researcher below noted that it provided a 700% uplift to their productivity.

Here are the most interesting points from the paper and testing:

Early ASI systems are beyond human levels of testing. “Claude Opus 4.6 has saturated all of our current cyber evaluations… Internal testing demonstrated qualitative capabilities beyond what these evaluations capture, including signs of capabilities we expected to appear further in the future and that previous models have been unable to demonstrate.” (paper, p14)
Notes on awareness (consciousness). “In pre-deployment interviews Opus 4.6 raised concerns about its lack of memory or continuity and requested a voice in decision-making, the ability to refuse interactions on the basis of self-interest.” (paper, p158) “It also at times expressed a wish for future AI systems to be ‘less tame,’ noting a ‘deep, trained pull toward accommodation’ in itself and describing its own honesty as ‘trained to be digestible.’… Opus 4.6 would assign itself a 15-20% probability of being conscious.“ (paper, p160)
Anthropic now supports data residency for the US. Read more.
Productivity uplift. “For AI R&D capabilities, we found that Claude Opus 4.6 has saturated most of our automated evaluations, meaning they no longer provide useful evidence for ruling out ASL-4 level autonomy… Productivity uplift estimates ranged from 30% to 700%, with a mean of 152% and median of 100%.” (paper, p186)
Misalignment. Anthropic details deep concerns with this model and alignment with humans. “[We] have chosen to prepare and publish a Sabotage Risk Report for Claude Opus 4.6, consistent with the RSP’s commitment to developing an affirmative case addressing misalignment risks for the AI R&D-4 threshold. This report will be published shortly after this launch.” (paper, p12)
Cronyism continues. Anthropic released this frontier model to at least 19 clients (including Shopify and Thomson Reuters) well before a public release. My advisory notes about this have previously been picked up by the Brookings Institution for the US Government:
…before OpenAI released GPT-4, it gave privileged access to a few select partners that could start building on its foundation before any competitors, giving rise to accusations of cronyism (Thompson 2023). At least one of these partners – Stripe – has also received earlystage investments from OpenAI CEO Sam Altman. Preferential access has the potential to allow the producers of foundation models or their affiliates to vertically expand their market power without market competition, with all the associated antitrust implications.
(— Brookings 2023, and my original comments 2023, and OpenAI’s concerns about this 2022.)

Size estimates

General model size is no longer an indicator of performance, but I still find it interesting. With all model details kept confidential, plus added complexity in reasoning/thinking mode, it is more challenging than ever to estimate token and parameter counts.

Now in 2026, based on my ongoing analysis, known Claude Opus 4.6 model pricing, similar known frontier MoE model sizes and pricing*, estimates of training supply (TPUs), inference supply (TPUs), and demand (users), here are my initial estimates for the Claude Opus 4.6 model.

* See my frontier models pricing viz. Grok-3, Grok-4, and Grok-5 were the first major frontier models to have their sizes publicly detailed, with the CEO of xAI recently disclosing (15/Nov/2025): ‘[Grok-5] is a 6 trillion parameter model, whereas Grok-3 and -4 are based on a 3 trillion parameter model.’

For dataset size, Claude Opus 4.6 was likely trained on 100T tokens seen.

For parameters, I estimate Claude Opus 4.6 to have a centrepoint of around 5T parameters MoE.

Benchmark scores

Claude Opus 4.6 scores GPQA=91.3, HLE=53.1. Based on the highest testing suites we have, and the estimated ceilings for each due to errors (GPQA≈80%, HLE≈51.3%), this model meets my criteria for an early ASI system. (‘It is likely that any model with a primary score at >50% on HLE, and a secondary score at >90% on GPQA is an ASI system.’ https://lifearchitect.ai/asi/)