Which AI tool do you use for documentation?

I received this table from a AI introduction course. It is the authors recommendation for the AI to the particular job.

This table is horrible, I have no clue where the authors got this data from, my guess is either astrology or hand reading.
Perplexity is not a model. Claude models can access 'internet' just fine. Cost of API use is also made up random data. There's also no such model as 'Claude' or 'Gemini', there are about a dozen models in each of those families with wildly varying API costs. (at the moment, the top Claude model is the most expensive out of all). Context window data is made up too, Claude models have significantly larger context windows than GPT 4o, yet they are both claimed to be 'Large', Gemini models generally have the same as Claude models, while one is shown as 'Large' and the other is 'Very Large'.

I'm confused. In one sentence you say Claude models can access "internet' just fine. Then two sentences later There's also no such model as 'Claude' or 'Gemini,'and the the follow-up statement, the top Claude model is the most expensive out of all. You do later say the the term of model families.

I'm not going to argue the merits of the table because my knowledge on the subject is so little I can't spell AI.

You believe the table is horrible. Your comments of where the data came from reminds me of some sites that I avoid visiting because of the acrimonious commentary.

I stumbled across a 15min video "Nano Tips for Using AI to Boost Productivity for Any Role with Rachel Woods.  Our public library has a Linkedin Learning site that patrons can use to find training material at no cost.

I've used the training material at the site to learn Inkscape, Gimp, Microsoft 365 and lately Canva. I spent more time watching the video than necessary because "I thought" the hosts was rather attractive. Sorry ladies part of my male brain has failed to develop and pretty still attracts me. The poster was offered as a free download.

I do note, missing from your response was any alternative material. An offer of alternative material might have gone a long way in helping me temper my response or avoid responding at all.

Claude is a model family, not a model. Claude model*s* can access the internet just fine, as in, all of the models in the Claude family can. It's not even related to the model itself actually, but the tooling around it.

'the top Claude model is the most expensive' also makes sense, the Claude model family has multiple models, and the top one, Opus (or Fable now) is by far the most expensive among all of these mentioned models/families.

I specifically didn't argue against the 'which model is best for what' thing, because it is heavily subjective. 

I could've included sources for pricing and context size, because those are hard facts, so here they are:
https://developers.openai.com/api/docs/models/all
https://ai.google.dev/gemini-api/docs/models
https://platform.claude.com/docs/en/about-claude/models/overview

I cannot offer any guidance on what model is best for what, because again, it's highly subjective. There is some general advice regarding what's currently the best for coding or creative writing, but it changes so quickly, and has so many factors to consider, that I cannot reliably make any good claims regarding it. My point is that the table shown above, and the vast majority of other comparisons like this are hilariously incorrect, even on parts where objective claims can be made. 

The general consensus is that Claude Opus and Claude Fable are the best for coding. What does best mean though? Fable is so ridiculously expensive that most people can't meaningfully use it. DeepSeek V4 Pro is definitely 'worse', but hundreds of times cheaper in practice. So which is better now? The one that will get something done in the least amount of prompts? The least cost? The least time? Is privacy a concern? These are not questions that can be narrowed down to a simple table.

As an analogy, think if someone made a chart about the best microcontrollers, and wrote some generic thing like "STM32 is the most expensive" and "RP2040 is the cheapest", or that "ESP32 is good for hobby projects" and "Renesas is good for automotive". These are extreme oversimplifications bordering on lies.

I didn't provide alternative material, because I cannot provide anything factual with no concrete information regarding the specific use case, just like how I couldn't recommend a microcontroller for someone who just says 'what microcontroller is good'. I would have suggestions for specific use cases or constraints, which I could almost certainly provide supporting sources for, although they would still be subjective. The goal of my comment was not to say that this table is bad and I know so much more, it was to show that you simply cannot factually provide good advice oversimplified to this degree. 

I didn't intend my response as an attack against you, you're obviously trying to help, I'm just trying to show how these AI guides are completely useless trash, I apologize if my comment came off as such.

Claude is a model family, not a model. Claude model*s* can access the internet just fine, as in, all of the models in the Claude family can. It's not even related to the model itself actually, but the tooling around it.

'the top Claude model is the most expensive' also makes sense, the Claude model family has multiple models, and the top one, Opus (or Fable now) is by far the most expensive among all of these mentioned models/families.

I specifically didn't argue against the 'which model is best for what' thing, because it is heavily subjective. 

I could've included sources for pricing and context size, because those are hard facts, so here they are:
https://developers.openai.com/api/docs/models/all
https://ai.google.dev/gemini-api/docs/models
https://platform.claude.com/docs/en/about-claude/models/overview

I cannot offer any guidance on what model is best for what, because again, it's highly subjective. There is some general advice regarding what's currently the best for coding or creative writing, but it changes so quickly, and has so many factors to consider, that I cannot reliably make any good claims regarding it. My point is that the table shown above, and the vast majority of other comparisons like this are hilariously incorrect, even on parts where objective claims can be made. 

The general consensus is that Claude Opus and Claude Fable are the best for coding. What does best mean though? Fable is so ridiculously expensive that most people can't meaningfully use it. DeepSeek V4 Pro is definitely 'worse', but hundreds of times cheaper in practice. So which is better now? The one that will get something done in the least amount of prompts? The least cost? The least time? Is privacy a concern? These are not questions that can be narrowed down to a simple table.

As an analogy, think if someone made a chart about the best microcontrollers, and wrote some generic thing like "STM32 is the most expensive" and "RP2040 is the cheapest", or that "ESP32 is good for hobby projects" and "Renesas is good for automotive". These are extreme oversimplifications bordering on lies.

I didn't provide alternative material, because I cannot provide anything factual with no concrete information regarding the specific use case, just like how I couldn't recommend a microcontroller for someone who just says 'what microcontroller is good'. I would have suggestions for specific use cases or constraints, which I could almost certainly provide supporting sources for, although they would still be subjective. The goal of my comment was not to say that this table is bad and I know so much more, it was to show that you simply cannot factually provide good advice oversimplified to this degree. 

I didn't intend my response as an attack against you, you're obviously trying to help, I'm just trying to show how these AI guides are completely useless trash, I apologize if my comment came off as such.