Opus 4.6 was working fine after the whole cache problems were solved. Now after the release of Opus 4.7, Anthropic has completely removed Opus 4.6. Why is Anthropic taking such poor descisions and screwing up with their customer base?
Generally it is a good practice to keep v.[N] and [N-1] in active deployment. But lately a lot of policies from Anthropic are confusing at best, and at worst plain subscriber-hostile.
If costs are drowning them, they should rather be a bit more honest about the messaging and hike prices of their lower ($20 / $100) tier by ~$10 and enforce stricter limits (optionally).
I liked 4.5 and 4.6, but none of them are available outside of using API now - and thats a bummer given the 2~2.5X overall costs
All good! I wish it was still in the model picker control too, would be much easier & clearer. But at least this way you can choose any Claude still available and turn off the 1 Million token window if you prefer.
It’s definitely an interesting decision to suddenly make it appear as though you can’t use it, offer it silently on the backend and push the higher usage new model.
I get these models are getting huge and costs are quickly adding up but at least be transparent. GitHub Copilot straight up said hey we’re not going to offer these anymore get Pro+ for 4.7. If nothing else you can use the overlap time to test and switch.
What alternatives are people using? I really liked Claude code up until a couple of weeks ago when it went a bit crap. I am a solo dev that works on my own projects as well as professional work. I am that person people think knows about computers. I'm not, I can make web pages and use a terminal.
The part of the entire "AI" - but not-artifical and not-intelligent - marketplace that I find amusingly curious is when any users expect the large model providers to take actions which are not 100% objectively for the purpose of resulting in more income from less computation resource.
Any action on the part of the model provider, which is not completely in the direction of more money for less work, is actually working against the interest of the provider, the shareholders, and various investors. Any such action would open the specific employees or parties taking that non-profit maximizing action to official review, reversal of action, and possible reduction in responsibility or termination of duties. That is completely within the job description of employment at any of the places, including Anthropic.
So, users, be informed, the model providers are not allowed to take user's best interest in to scope for making utility decisions.
The company will at it's own discretion; reduce customers' allowance of capability, reduce the computational cost for performing the same actions at a later date, reduce the precision and utility of results iteratively as the number of users increases overall demand. Model providers will showcase higher performance and precision to non-paying 'potential customers' and then nominally deliver increasingly reduced performance up-to the point of a specific and acceptable ratio of account cancellations.
This is specifically in order for the model providers to extract as much coin for as little work as possible: that is the purpose of the many profitability matrices and customer load balancing 'knobs' that are run using the very best models. It is a given and very reasonable to expect an explicit requirement from the top that the internally facing model capacity is NEVER degraded for the tasks of increasing revenue per unit of work; that is, to make as much as they can at any point in time, and make more at every point in time than they did before.
The responsible use of these models should be; use the best performance (i.e. the free tier) to assist in the development of infrastructure to run models which are actually under your control; either on in-house hardware, or on leased VPS instances with specific hardware and performance guarantees. Then use models under one's own control to develop stable and irrevocably performant workflows.
Any reliance on public facing subscription models for actual work is 100% guaranteed to deteriorate in functionality over time and waste money trying to maintain performance as measured by cost per token over time. That is the fact of the matter.
The casino increased all their fees, reduced their limits and want more from your credits.
The owner of this casino (Anthropic) decided to upgrade the slot machine (Claude Code) with a new random number generator engine (Opus 4.7) and decommissioned the old one (Opus 4.6).
They claim that the new engine (Opus 4.7) gives better odds and generates results faster with every press of the button, with a catch that it takes more credits on every turn.
i dont think you need to make it malicious, this is what not having enough compute looks like in practice.
its true of gpt5, its true of why antigravity rate limits, and its true of claude.
there is no possible action thats not user hostile when you dont have enough compute. You either reject customers which is hostile and kills your companies growth or you do the best you can with what you have.
its not like these companies under built, they're building as fast as they can.
I mean if a corps goals dont align with their customers (which almost universally they do not), its pretty safe to assume every action is malicious. Anthropic need to generate profit from a service where users burn cash like it's the weimar republic, every choice they make will be to the detriment of the user up until users stop paying.
If costs are drowning them, they should rather be a bit more honest about the messaging and hike prices of their lower ($20 / $100) tier by ~$10 and enforce stricter limits (optionally).
I liked 4.5 and 4.6, but none of them are available outside of using API now - and thats a bummer given the 2~2.5X overall costs