Claude Sonnet 4.6 Model Brings 'Much-Improved Coding Skills', Upgraded Free Tier 44
Anthropic has released Claude Sonnet 4.6, the first upgrade to its mid-tier AI model since version 4.5 arrived in September 2025. The new model features a "1M token context window" and delivers a "full upgrade of the model's skills across coding, computer use, long-context reasoning, agent planning, knowledge work, and design." From Anthropic: Sonnet 4.6 brings much-improved coding skills to more of our users. Improvements in consistency, instruction following, and more have made developers with early access prefer Sonnet 4.6 to its predecessor by a wide margin. They often even prefer it to our smartest model from November 2025, Claude Opus 4.5.
Performance that would have previously required reaching for an Opus-class model -- including on real-world, economically valuable office tasks -- is now available with Sonnet 4.6. The model also shows a major improvement in computer use skills compared to prior Sonnet models. The free tier now uses Sonnet 4.6 by default and with "file creation, connectors, skills, and compaction" included.
Performance that would have previously required reaching for an Opus-class model -- including on real-world, economically valuable office tasks -- is now available with Sonnet 4.6. The model also shows a major improvement in computer use skills compared to prior Sonnet models. The free tier now uses Sonnet 4.6 by default and with "file creation, connectors, skills, and compaction" included.
Iffy (Score:2)
Coding win and what else? (Score:2)
Excluding writing software, what AI products are bringing billions in revenue? Are there any billion+ revenue products outside of writing code?
That is a billion in revenue and not a billion in company or product valuation.
It's a billion in yearly revenue today and not an extrapolation of 500% monthly subscriber growth for the next 4 years.
Re: (Score:2)
That is the "much improved" factor: More errors for your money and time!
More is always better, right?
Ah, how quaint! (Score:2)
The incessant "WoM advertising" campaigns of the last days of the dotcom bubble...
Pepperidge Farm Remembers.
Re: (Score:2)
Greenspan is known for inflating bubbles, not for deflating them, sorry.
Re: (Score:2)
What actions did Greenspan take that would have "popped" the dotcom bubble?
Re: (Score:2)
He took a needle and stuck it in the bubble. Da doy!
Re: (Score:2)
What actions did Greenspan take that would have "popped" the dotcom bubble?
It's only been 30 years since his "irrational exuberance" [wikipedia.org] speech. How soon we forget.
Re: (Score:2)
Okay, that's cool, but what else besides the speech? Was that it?
compaction is an anti-feature (Score:2)
And besides I've had to deal with compaction in sonnet 4.5, too. So, fake news all the way down, I think.
Still, sonnet is pretty good at a lot of tasks.
Re: (Score:2)
My impression is that compaction stems from (somewhat desperate by now) attempts to bring LLM effort down. Is that somewhat accurate?
Re: (Score:2)
yes. However the results are almost always brain dead. it's far better to make your own continuation prompt.
Re: (Score:2)
Interesting.
Hope it's faster than Sonnet 4.5 (Score:3)
Hope it's faster than Sonnet 4.5. For some reason, SQL Server Management Studio *insists* on using this model as the default, even if you chose a different default. Every time you open a new chat, it reverts back to Claude Sonnet 4.5. And I always know when this happens, because it is so SLOW to come back with responses. By contrast, GPT-4o, which is quite capable for most tasks, is quite fast.
Re: (Score:2)
Why does a DB management system need to use AI in the first place? What are you there for, just to double click on the desktop icon to start it and it does the rest itself?
Re: Hope it's faster than Sonnet 4.5 (Score:2)
Hi AI. I want you to SELECT data FROM the table users WHERE the hire date is today.
See how much easier that is than writing sql? ðY
Re: (Score:2)
It's a valid question.
Not all of SQL is simple. Like any language, there are corners that aren't often touched. For example:
- Common table expressions (CTEs)
- OUTPUT clauses
- CROSS APPLY
- Extracting data elements from a JSON or XML field--the syntax for these is often arcane and not well known
AI is a big help writing or modifying such queries.
Then there are queries that look into the metadata of the database. For example:
- AI can examine the indexes on a table, explain what they optimize, and propose method
Re: (Score:2)
> - Or, faced with an old 1,000 line stored procedure that's been there forever and nobody knows exactly what it does, it can summarize and analyze it, and refactor it for maintainability.
If no one knows exactly what it does, how does anyone know if the refactored version does the same thing?
Re: (Score:2)
Is this a trick question? I would do it through testing.
Once you know what it's supposed to do, and what its various outputs and side effects are, those can be measured and tested compared to the original, or corrected, as the case may be.
Re: (Score:2)
Ah, sweet summer child. For a complex enough piece of code you won't know everything it does and all the side effects until it does them. I'd say a 1000 line stored proc falls into that category.
Re: (Score:2)
Oh I'm well aware. I've refactored plenty of those into smaller, more manageable stored procs, that are testable. The thing is, everybody is afraid to touch those huge, ancient stored procedures because they are afraid they'll break it. The truth is, they were already broken in 100 ways, just nobody knew it. So taking the risk to refactor is always well worth the risk that you might miss one thing. And when you find that one thing you missed, you'll be able to fix it after the refactor.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
The next version will definitely require the interactive chair attachment. Figuratively fu****** the user is just not enough anymore as LLMs get better and better!
Re: (Score:2)
Re: (Score:2)
4o is gone now! https://www.businessinsider.com/openai-retires-gpt-4o-user-backlash-chatgpt-ai-2026-2 [businessinsider.com]
Re: (Score:2)
Yeah I'm one of the people who isn't happy about this. But not for emotional attachment reasons, but because it's "good enough" and *FAST*. It's faster than almost all the other models I've tried.
This is all so stupid (Score:2)
On the one hand we get told how their stupid almost-random world-generator will replace everyone, on the other hand all their tools
Still do not fulfill the most basic requirement any tool has to fulfill: reliable and dependably do the job. Literally screwdrivers are more useful.
Re:This is all so stupid (Score:4, Funny)
Re: (Score:3)
Sorry to disappoint you, but Liz Truss beat you to it.
Re: (Score:2)
That was the ultimate power move from the Queen.
Re: (Score:2)
+1, insightful.
Re: (Score:2)
What matters is that LLMs reliably and dependably say how fucking awesome I am
LOL.
FYI, if that's not actually what you want, it's fairly easy to fix. All of the models allow you to specify a "personal preferences" prompt that is automatically applied to all conversations. For example, my preferences prompt for Claude says:
Re: (Score:2)
Re: (Score:2)
Yes, and now it says to you, "I'm going to give you the unvarnished, cold, hard truth."
I wish. It's still a "people pleaser", just less so.
Re: (Score:2)
Re: (Score:2)
It just depends on the model.
I use Claude, mostly. Which one is most likely to give me the unvarnished, cold, hard truth?
Re: (Score:2)
Maybe I'm easily impressed... (Score:2)
I am not associated with Anthropic or Claude beyond being a very recent paying user. It might sound like I'm shilling, I'm really, really not. I'm just surprised to be this excited by an LLM.
Claude is the most capable coding LLM I've ever used. I paid for it after trialling it for just a couple of hours. I am an AI - and particularly LLM - sceptic. I had used ChatGPT and Copilot and dabbled very briefly with Gemini and thought they were fine as Google replacements/augments, but they suck at coding, especial
Re: (Score:2)
More infinite monkeys? (Score:2)
Wanna buy a bridge?