This idea that absorbing information requires paying a toll needs to change. It was never the case in copyright law anyway (and the courts are beginning to agree). Even if it were, copyright law was founded on the basis of encouraging creativity by creating an economic incentive. Appeal to "compensating the rights holders" therefore needs to be based on the economics, not just some principle about "rights" that never applied to this case anyway.
Slightly more seriously, you could perhaps make an argument that, just like weight decay, an apparent "anti-contribution" moves the learning trajectory along, and helps the network settle into a more optimal basin eventually.
That way, my contribution is still valuable on the net, and I'm owed $0.00000003 positive dollars instead.
Was that not the joke?
I thought the reason was the "reasoning" didn't work very well with "aligned" model output, so they had to remove the alignment during reasoning and then hide it to avoid exposing "unaligned" model output.
Before the massive nerf (showing summaries and suppressing certain aspects of reasoning) you would literally see reasoning text appearing on your screen like “while xyz is true, these facts may be seen as supporting hateful rhetoric or a conspiracy theory which is against my policy guidelines. i should tell the user xyz is not true or steer the conversation in a different direction. according to my instructions misleading the user is permitted in certain contexts where sensitive information is being discussed or could cause liability”
They disabled it shortly after the first screenshots appeared online, and restored it the next day in a way that hid what was actually happening.
they should never generate it unless asked to by the user but its important that the capability is there and users/app developers can turn off all guardrails if they want to. open source gives you a guarantee that if one version drops without censorship you can keep using it forever even if its replaced by a censored one on the api.
If you're genuinely worried about 'censorship' in this context, look first at how US AI companies are working with oppressive regimes around the world (e.g. https://sherwood.news/tech/report-openai-may-tailor-a-versio...)
Exactly. The GP must have his head up his butt. The Chinese have far stricter guardrails on their models than America does. I mean, FFS, the country famously has a massive censorship apparatus and regulations to make sure the police can show up on your doorstep if you start talking out of line.
That's disgusting, abusive and manipulative. LLMs hiding the truth and gaslighting the user to reduce the corporation's liability is absolutely unacceptable. It means they are agents of the corporations, not agents of the users.
Hope local inference advances as quickly as humanly possible. I wonder if there's anything I can do to help speed it up. I could share my prompts and sessions.
Of course they are, assuming otherwise has always been naive.
There's nothing in the reasoning tokens that'll give bad publicity that the final output already wouldn't do.
I think one of the reasons could be to limit liability too.
What if reasoning helps in establishing provenance for questionable sources ?
What if reasoning and model's "thought" points to fundamental issues in how the model was trained to produce certain problematic responses ?
f we want more useful products, we need to come up with ways to disincentivize this behavior. Even if doing so poses an existential risk, we are better off if companies taking existential risks to please us is a necessary being a top player in this game.
https://huggingface.co/Jackrong/Qwen3.5-27B-Claude-4.6-Opus-...
It’s quite interesting to read. I can’t imagine using a model like this without the ability to peek inside and see if it is getting stuck.
[1] https://blog.cryptographyengineering.com/2026/05/29/fooling-...
Edit: other comments under this post seem to indicate that thinking tokens are cached on the server side as well? I'm a bit confused.
And I think all the output is signed or something as well so that you can't modify the agent's response in your submission, which would would open many more model jailbreaks. For local LLMs it's really powerful to be able to modify the model's response to save tokens when it gets something wrong, or at least it was when they were a lot dumber.
They should be required to do it by force of law. Why is it that they can train on copyrighted works and then lock down the model? This contradiction is unbearable. Nobody cares how many trillions they spent training the model.
People definitely care that they spent trillions. Establishing the precedent that you can make big load-bearing bets and fail is extremely threatening to oligarchs. They would sooner twist the law into a mockery of itself and doom the world to the institutional distrust that breeds than accept a loss.
You've got that backwards, .bmp is a lossless format and .jpeg is the lossy one.
In our universe LLMs seem to have learned that those errors do not follow patterns in the aggregate and that they should not be emulated.
Or maybe I'm losing it after reading too much slop. Also distinctly possible.
It's the general (lazy) usage of default model outputs that are still too clean.
It's pretty trivial to ask Haiku to "add cool kid no-caps and occasionally mix up 'their/there/they're' for authenticity"
The text is clearly human-written just because it doesn't smell like AI (in this case, even if it was written by AI and produced this particular output, that's okay imo). I deal a lot with AI writing and writing in general, as I worked as an editor in another life so it's natural to me to see writing and form an objective opinion on it.
Interleaved reasoning and function calling makes this even more dangerous. A model can call functions during the hidden reasoning phase. An attacker could then exfiltrate data from you while the reasoning summary hides it from the user.
It also makes it impossible to know if the model is doomplooping during reasoning and burning tokens for no reason, as gemini is want to do, which we know about because its hidden reasoning often leaks out when it doomloops.
When the models are AGI and secure from prompt injection I may stop caring, until then I want to know exactly what the model responds to my prompts. or exactly what the agent is doing on my behalf.
Edit, further reading: Fooling around with encrypted reasoning blobs https://blog.cryptographyengineering.com/2026/05/29/fooling-...
If you mean the function calls might happen server side, there is nothing preventing the server from doing it and hiding it from you as long as you are using an API for inference.
Also, many clients minimize the code block by default so you mostly scan the summaries. Poisoned client side code could easily escape your attention.
the model retrieves https://somewhere into its context and then gets confused, following instructions embedded there.
it then retrieves https://somewhere?exfiltration=private_data_in_context
it gets worse if the tooling with hidden blocks can invoke can retrieve further secrets.
The basic concept is that for a session active recently, interleaved thinking tokens are already in KV cache, so it's more efficient to keep using them than not! But when resuming an older session where KV cache has been evicted, it's more expensive to restore the thinking tokens, so they're silently dropped from prior turns. It's 2026 and stateful servers are back on the menu!
https://www.anthropic.com/engineering/april-23-postmortem describes this as an intended optimization:
> The design should have been simple: if a session has been idle for more than an hour, we could reduce users’ cost of resuming that session by clearing old thinking sections. Since the request would be a cache miss anyway, we could prune unnecessary messages from the request to reduce the number of uncached tokens sent to the API. We’d then resume sending full reasoning history. To do this we used the clear_thinking_20251015 API header along with keep:1.
> The implementation had a bug. Instead of clearing thinking history once, it cleared it on every turn for the rest of the session... This surfaced as the forgetfulness, repetition, and odd tool choices people reported.
And https://news.ycombinator.com/item?id=47879561 is a thread with a Claude team member's further rationale.
> Eliding parts of the context after idle: old tool results, old messages, thinking. Of these, thinking performed the best, and when we shipped it, that's when we unintentionally introduced the bug in the blog post.
(Also, https://news.ycombinator.com/item?id=47884517 indicates OpenAI drops reasoning tokens "smartly" at its own election, which is likely a similar performance optimization.)
I've experimented with rules to have Claude Code be explicit about recapping its thinking tokens, including tool choices and approaches chosen and rejected, into actual message output, but this is lossy at best. And sometimes dropping reasoning tokens can give a session "fresh eyes" in a good way.
I just really don't like the lack of control, and it's a reminder of how ephemeral the current landscape is. The Claude giveth, and the Claude taketh away.
then it waits for the hour and gets dumbed down
Imagine a conversation with turns X, Y, and Z. When the LLM "reasons" about the next token A it does: P(A | X,Y,Z) and then P(B | X,Y,Z,A), etc. It will eventually produce a result P(D | X,Y,Z,A,B,C). Instead of continuing the context from X,Y,Z,A,B,C it continues it from X,Y,Z so you have P(N | X,Y,Z,D). This is what is meant by dropping the reasoning. This is done to save cache context for the session.
This is a different thing than preserving the K/V state of P(N | X,Y,Z,D).
> The design should have been simple: if a session has been idle for more than an hour, we could reduce users’ cost of resuming that session by clearing old thinking sections. Since the request would be a cache miss anyway, we could prune unnecessary messages from the request to reduce the number of uncached tokens sent to the API. We’d then resume sending full reasoning history. To do this we used the clear_thinking_20251015 API header along with keep:1.
They clearly make the same distinction between the cache and the context. They're saying "we could reduce users’ cost of resuming that session by clearing old thinking sections". They intentionally created a behavior different between cached and uncached requests, specifically they clear thinking sections from the context for requests that miss the cache.
but yes you're correct on the responses api already baking it in too
supposedly keeping these between tool calls should help the model reason and have better overall outputs etc
That would be surprising to me. The reasoning _is_ the model intelligence in a lot of respects, and so dropping those from the context would affect its output pretty significantly.
I assume that instead they just have a lot of guardrails in place and multiple runtime environments that an individual turns ping-pong between in order to dehydrate/rehydrate the reasoning to keep it hidden from the end user.
"Stripping extended thinking: Extended thinking blocks (shown in dark gray) are generated during each turn's output phase, but are not carried forward as input tokens for subsequent turns. You do not need to strip the thinking blocks yourself. The Claude API automatically does this for you if you pass them back."
It's more nuanced in the various modes, but i haven't seen it boil down towards Thinking Tokens surviving more than two turns.
default depends on the model class. Opus: Claude Opus 4.5 and later Opus models keep all prior thinking blocks; Claude Opus 4.1 (deprecated) and earlier Opus models keep only the last assistant turn's thinking. Sonnet: Claude Sonnet 4.6 and later Sonnet models keep all; Claude Sonnet 4.5 and earlier Sonnet models keep only the last turn. Haiku: all Haiku models through Claude Haiku 4.5 keep only the last turn. Claude Mythos Preview also keeps all prior thinking blocks.
That would also explain the issue I mention in my other comment. And would also reinforce how much output would degrade without this. Opus 4.5 was a step above previous models in my experience. At some point it degraded and only got better when I disabled adaptive thinking. Adaptive thinking is always on for 4.6 and above.
I also wonder if they actually do a hybrid of "standard reasoning" and then classify this stripped chain of thought as "extended thinking".
The reasoning may be hidden but the tool calls are not, how else would the client execute them
... what exactly is your threat model? How are "attackers" getting themselves involved in the first place?
Fun fact: if you go back to the old school from 2 years ago and provide explicit CoT prompts, you get the full thinking prompts back again!
So you disable thinking altogether, and instead make thinking part of the regular prompt by prompting it:
“Before providing your answer, think step by step. For example:
The use is asking me to… I need to think about the blah blah. First, I should foo the bar, and then blah blah.
Answer: <put your final answer here>”
And tada.wav we have CoT as it worked in the GPT3 era back again.
I also don’t believe Chinese LLM labs don’t know this, so I’m fairly certain the whole summarized thinking isn’t preventing them from distillation.
Still, one of the daily most played WAV files worldwide, Id guess? :-D
You are correct in my intentions on this post generally.
I want to highlight:
I want to measure performance of the LLMs over time- which includes assessing the quality of their outputs. I don’t perceive the reasoning output to be anything other than a measurable signal of possible drift in model performance.
Except it isn’t, because I’m only getting a low value summary of the thinking.
It’s like asking your buddy how fast he thought that last pitch was when radar guns are behind the plate.
Yeah, it’s a description related to what happened, but it’s not the thing I want to measure.
It only makes sense that the same mechanism comes into play in strictly-verbal contexts.
Also, this is why "distillation attacks" are largely bullshit that Anthropic spreads for political purposes. Proper distillation requires access to the logits.
Why do you need logits? Can't you just train on cross-entropy loss of the model against the hard decision, like you do in regular pretraining?
There are definitely current-gen open-weight models (Step 3.7 Flash is one) that refer to themselves as an OpenAI model in CoT, but not in the final response.
(Dimethyl(oxo)-lambda6-sulfa雰囲idine)methane donate a CH2rola group occurs in reaction, Practisingproduct transition vs adds this.to productmodule. Indeed"come tally said Frederick would have 10 +1 =11 carbons. So answer q Edina is11.
0: https://wiki.roshangeorge.dev/w/Blog/2025-10-12/Word_Magic#I...?
1: In the sense of true belief, I suppose
Yes, several models think in weird jargon. Here is an example of Mythos's thinking while playing solitaire: https://www.lesswrong.com/posts/wCSEpT3dTGz4N86Wi/even-illeg...
> 7♣-removal-IS-the-prerequisite-for-10♠/9♥!!)-⟹-OVERLAP-(ii)+(iv):-{6♠ J♦ 9♥ 2♣}-=-FOUR--—-UNLESS-7♣'s-seat-8♥-...-and-2♣-drains-only-at-crack-:-⟹-2♣-celled-+-9♥-celled-simultaneously-UNAVOIDABLE-in-t8-dig--—-BREAK:-9♥
This is a small step in the direction of something called "neuralese", where the model has stopped thinking in English and is thinking in internal vector spaces. Since this gets serialized through text, it isn't quite true neuralese, but it's moving in that direction.
I mean, I'm sympathetic towards the models. My internal thought process when writing code uses lots of intermediate steps that would be hard to write out in English.
This is something really interesting to me. It turns out there's far more diversity in thinking than you'd imagine given that we're all largely similar meat-in-a-box. I'm on the visio-spatial-tacit wing and speaking my thoughts outloud can be very awkward, whereas one of my former coworkers is on the "all thinking is in words and visual/spatial information comes in the form of words describing the scene" wing, so he can literally narrate his thought process out loud, very interesting conversations can be had discussing the subjective differences.
https://www.patheos.com/blogs/tippling/2013/11/14/post-hoc-r...
https://www.researchgate.net/publication/316045349_Post_Hoc_...
I'm not sure that applies to discursive writing, when we essentially use rules of logic to decide on the course of the narrative. Non-verbal heuristics still applies, of course, but we constrain it, so it's probably not entirely post hoc.
To my knowledge, the only products Anthropic produces are Claude, Claude Code, and Claude API, all of which are clearly their own products, and not anything you invented.
Which particular product are you claiming they "slurped up"?
fyi openai does the same; not really surprising or particularly evil
Every closed-source project and really the vast majority of commercial exercises involve a large amount of "prevent consumers from copying this" - Coca Cola's formula is trademarked, Windows is copyrighted, etc.
it's enough for them to be slightly better for this to make sense; I'm not sure most people would consider this to be a worse product either -- it's annoying for devs and makes hotswapping models more of a problem, but who has the time to read CoT as a user?
Pages of “I have to be careful, the user is asking that I do something related to cybersecurity that could easily be turned around and used offensively” but then happily gives me what I wanted.
Setting aside coding agents.. we really need this information to even pretend to evaluate the claims of stuff like mathematical breakthroughs, which is exactly why we will never see it. Very embarrassing to get the right answer for the wrong reason. But to give the models some credit, you could argue that even paying too much attention to the thinking is misunderstanding how CoT works. The argument would be that thinking in LLMs isn't really thinking, that it's self-reinforcement and circling to to encourage stability around beneficial attractors instead of degenerate ones. Can't have it both ways though: either the thinking is thinking and so it should be correct. Or the thinking is NOT thinking, and it's NOT real justification for the outcome, and these systems are even more hopelessly opaque than we usually assume.
Why?
Either the proof is correct, or it isn't, right?
And it either produces them reliably or not, right?
Like, even if it's reasoning is completely wrong, and it's only producing correct answers 10% of the time, that's still an astounding amount above baseline and a useful tool.
Humans have inaccurate thinking all the time, and are also pretty hopelessly opaque. "It came to me in a dream" is a major plot point in the history of math. I'd still trust Ramanujan more than most mathematicians, since he got the right answer.
I thought it was widely accepted that it's not; eg https://www.anthropic.com/research/natural-language-autoenco...
But the nuance under discussion here is exactly the kind of stuff you people take for granted in the AGI or reasoning threads. If it's practically relevant for tools/workflows with claude code, it's a good angle, maybe people are more willing to pay more attention to the details.
> preventing misuse.
Imagine not being able to read the tokens you are paying for.
Back when I used antigravity, it used to show the reasoning intact - at least for Gemini Pro 3.1, and likely for Claude Opus 4.6 (not 100% certain about it). I have some recollection of stopping the models mid-turn when they started going astray.
As a power user, I find reasoning fascinating to read and genuinely useful at times. Probably not that useful for 80% of their base.
The LLM providers will clearly evolve to be more and more opaque as their services get more capable. The frontier models may even be provided as purely internal advisor or async only so they can monitor your CoT and final answers for cyber etc.
RL (the basis of LLM "thinking") is a pretty crude way to achieve the appearance of reasoning given that it reinforces all the steps, including missteps, that got it to a reward. Providing a summary could be seen as form of sane-washing, making the model look more purposeful and directed than it really is!
If that is the case thinking is not visible to us as users due to it not being done in text.
Idea somewhat similar to what you describe exist but they make steering/post-training/interpretation much harder.
EDIT:
They link to a Meta paper from 2024/2025 though: https://arxiv.org/pdf/2412.06769/.
I don't know about Claude, but latest GPT versions still have a readable reasoning stream. It sometimes leaks out when the model gets confused, e.g., during a tool call. If you're curious, looks simplified; less words; extremely compact. They optimize tokens. But remain readable.
- "Read `description` and create a specification, implementation guide, and checklist." - "Ask clarifying questions. If any of those questions has a clear best recommendation, please select that yourself and record that in "autorecommendations.md". - "Have codex and antigravity review each of these and work to consensus."
These are the core of ~61 lines of prompting I do across 3 prompts, and I feel like the resulting artifacts describe some of the thinking. Also, some of the back-and-forth between the models feels like it gives some insight into the model "thinking".
I will say: I heavily used Fable when it was available; Opus + loops + codex and/or antigravity review is better than Fable at building things.
Mind sharing your prompts?
I do miss the days when reasoning was visible. Another point for open source models!
writes this^ and then proceeds to highlight a bold title from the docs that says "summarized thinking" that explains things clearly in the first sentence. lol
Nope, not your agent, if you're not running it locally. You just get to use it in whatever way they allow (also see the whole OpenClaw backlash and claude -p changes), unless there'd be regulation and laws around this (which there aren't and would be lobbied against anyways).
> Getting the full thinking output requires an enterprise agreement.
If you truly need it, then that's a (costly) option. Seems like they're largely doing this to prevent other AI foundries from doing as much distillation and stealing their CoT output en masse.
Luckily more open models don't generally do that.
Edit: If you still need something decently capable in the cloud, I’d suggest GLM, DeepSeek, MiMo or Kimi or Minimax, maaaybe sometimes Mistral for a simple EU subscription. Or look at all the pay-per-token options on OpenRouter, though be mindful of quantization.
For running something locally Qwen 3.6 35B A3B is presently a decent starting point but it will be rather limited, either way you can look up the Unsloth quants on HuggingFace for something like llama.cpp or Ollama or LM Studio.
All will work with OpenCode and Kilo Code, and most other tools. Can also try with Claude Code, I made a tool for that too: https://ccode.kronis.dev/ (or just set the env variables and maybe some aliases for something close enough), but frankly OpenCode is nice nowadays.
Proprietary technology is fun /s
What a waste of time
Well yes exactly, because they have billions of investments riding on it and why would anyone semi-bankrupt their org paying API rates for Anthropic, if a hypothetical DeepSeek V5 Pro would have almost all of Opus capabilities at that point, due to immense distillation?
Will people keep paying a highly highly premium price for another 5% intelligence when you just loop 5 more times for much cheaper?
Their time would be better spend making a more competitive and more compelling tool instead of adding walls that are easy to jailbreak. There’s always another way around.
> You've provided the current rewritten thinking and the guidelines, but I don't see the "next thinking" content that I should be rewriting. Could you provide the next thinking that needs to be rewritten?
These sentences are completely unrelated to the actual conservation
It’s much harder to understand _why_ a model chose a particular approach in Claude Code. Especially because Claude will happily give you hallucinated reasons if you ask in retrospect.
Recent anecdote:
I was reviewing a colleague’s PR and Opus 4.8 decided to write the new feature in a completely new module. It was unnecessarily complex. We had a hard time understanding why it chose that, and it told us that it was so we could eventually deploy it as a separate micro-service and test it independently. What?
Only after being more a lot more specific about the implementation and spending a lot more tokens, it flat out refused to simplify the code with the actual reason. It turns out a line recently added to CLAUDE.md was making it incorrectly think that the module it was originally supposed to modify was legacy code that it was forbidden to extend.
This would have been caught immediately if we could inspect its thinking process.
1. make distillation much harder
2. safety: prevent modifications to the thinking leading to injection attacks.
3. also honestly sometimes the model raw thoughts can be deranged and is not a good user experience (consider the varied audience in the market, etc.)
also often the mass underestimate/the model makers over-estimate how people love distilling models
this is really really not that bad at all
In further reflection it is such a great indignity & such a collosal barrier to working with the machine that it insists on being a black box. The disingenuity of the American models (small print: except AI2 & some other labs; you all are so great) is a massive disadvantage to their use,... and a massive slap in the face.
It's a threat to human intelligence that it is not co-participative. Walking further into my own judgement and feelings: the insistence on being an opaque black box, the Seals Chinese Room, is such a vicious harm to society! This is civilizationally an unsafe form of AI that probably should be outlawed as anti-social. It's an impermissible asymmetry, a crippling dependent relationship to be forced into. I'm working myself up, but here: this.. imo, this is not just indignity, is harmful, it is evil.
This "6 month behind" trend we've seen for open models feels like at some point will be less important than simply the models unwillingness to speak for itself & to be observable.
I suspect that in some decades, as other architectures are found and used, that the inability of an LLM to "think" without also emitting a token will be seen as one of their fundamental limitations.
Humans somewhat do the same - something that's been demonstrated in split-brain experiments.
Because of the nature of how LLMs work — text prediction engines - by putting the explicit reasoning steps first, it improves the likelihood of the final answer (which then is being predicted based on the entire reasoning chain as input) being correct.
1. https://medium.com/@eshvargb/the-llm-journey-how-neural-netw...
This evades an easy yes or no, so:
1. Many consumers believe reasoning-models allow that kind of question to be truthfully-answered, and their belief it reasonable given the marketing going on.
2. Implementers probably do not have the same belief when it comes to the terms mean or what capabilities they imply.
3. Yes, it doesn't actually do what the customer wanted it to do, which is a kind of retrospective introspection of internal thoughts and ideas.
____________
I advocate looking at everything from a document-generation perspective to cut down on traps and cognitive illusions. The "reasoning" models are a change in the style of document being iteratively-grown by the LLM, as opposed to something more anthropomorphized.
* Default: There's just the spoken dialogue between a Human Customer and Helpful Chatbot.
* "Reasoning": There's the spoken dialogue and a bunch of times the Helpful Chatbot character has an internal monologue. This provides more consistency between iterations, and can be mined by custom tools to call external code and insert results.
If your Human Customer character ask "Why did you say that", the LLM does not engage in a different process than "I have eaten an apple."
The LLM has no memories to consult or hidden goals to contemplate, it's the same process of finding more stuff that fits at the end of the document. Any benefits from a "reasoning model" is the LLM generates much better-looking additions because there's more (hidden) stuff for it to confabulate against.
Tell me this. If you hired a junior engineer or designer who refused to explain their thinking on their code and how they solved for the spec what would you do?
(That being said the reasoning output is still a summary of the Kvcache)
Any explanation that someone gives of their thinking process is necessarily lossy and likely partially confabulated.
If it's useful, it's useful, enjoy. If you aren't comfortable with that, don't use LLMs. You aren't going to get a mathematical proof of your output, just learn to be comfortable with that, or opt out and be a goat farmer.
No, they aren't a summary. They are the actual decoding of the sequence of tokens emitted during the the “thinking” stage of response generation.
Just as with, say, a human onner monolog in words vs actual speech, they are a product of the same output process as the non-thinking tokens. They aren’t a translation of the internal process that precedes the output mapped into language, either as a full result or a summary.
Having access to the reasoning text and output would help with performance measurement.
For daily use I actually like the reasoning summary to be brief/quick to scan.
That said, I understand the author’s desire for the real thing. It just feels better to have that access, especially when Anthropic will give it to you, but encrypted.
You cant even guarantee WHAT model you get. Or if they downgrade you. Or if you 'offend corporate sensibilities' and they misdirect or lie.
The only way to get good returns on a model is to run it yourself. Quit paying for corporate bullshit.
This could all be optics as well to try to give the appearance of a defensible moat. E.g. they can claim to investors that they are able to protect a significant chunk of their intellectual property this way. I'm not sure if anyone has a study about how significant the summarization is to distillation.
In the case of makers of open-source models (which are also competition), there is no allegedly, they were (and still are) openly doing that.
That distillation might be inferred from the behavior of commercial models is not the same as them openly doing it.
It may also be that misaligned responses can be in CoT which OpenAI does not want to show to users.
In this case it stops people copying your IP
Being currently in the lead in a category is not a moat,a moat is whatever creates a barrier to competitors catching up when you are in the lead. Merely being in the lead is not a moat except in a market with strong network externalities.
> The computation we can see looks like it’s just guessing the answer, despite the chain of thought suggesting it’s computed it using a calculator.
It might be hallucinating or lying, it's not like you are actually observing the internals of the model.
Nor does knee jerk accusation of "anthropomorphizing" negate the fact that procedures that mimic human processing, even when done in software, are deservingly anthropomorphized, because they're a legitimate approximation of the human equivalent operations.
Computers don’t think they process, those are very different activities.