So 1,5 days with cursor ultra and sonnet 4.5, its nice but a real developer is cheaper then this
1 Like
I suggest to use Claude, preferably thinking mode. you don’t need to pay separately. try it one month and see what happens. go for Ultra plan.
Thank you very much for the valuable information and opinion from your experience. I will get an ultra for 200 dollars a month and from there I will use Claude 4.5 in thinking mode and module by module I will start building my system. And by December there will be more updates which will surely help me as a person who is starting out. Thanks again.
1 Like
Thanks for the reply @condor .
According to the Dashboard, yesterday I used 11 Sonnet 4.5 requests, and 1 Sonnet 4.5 Thinking request. And yet I was charged about 40 requests total
Are you suggesting claude code? the CLI for me is a bit annoying i like how cursor/github copilot just use tooling inside the IDE. is claude code works the same in vs code?
once i tried gpt codex in vs code, it was a mess, it tried doing things with powershell and the python choking on itself, i don’t want that i want full integreation.
what do you suggest?
Roo Code extension
Can you clarify this for me please:
If I use Sonnet 4.5 in the chat, do I have the same capabilities described in the documentation for the API? I’m confused if using Cursor Agent to call the model counts as an API call. Do I need to configure anything or are these features only accessible via BYOK or something?
For example:
Claude Sonnet 4.5 introduces major advances in agent capabilities:
* **Extended autonomous operation**: Sonnet 4.5 can work independently for hours while maintaining clarity and focus on incremental progress. The model makes steady advances on a few tasks at a time rather than attempting everything at once. It provides fact-based progress updates that accurately reflect what has been accomplished.
Context awareness: Claude now tracks its token usage throughout conversations, receiving updates after each tool call. This awareness helps prevent premature task abandonment and enables more effective execution on long-running tasks. See [Context awareness](/en/docs/build-with-claude/context-windows#context-awareness-in-claude-sonnet-4-5) for technical details and [prompting guidance](/en/docs/build-with-claude/prompt-engineering/claude-4-best-practices#context-awareness-and-multi-window-workflows).
Enhanced tool usage: The model more effectively uses parallel tool calls, firing off multiple speculative searches simultaneously during research and reading several files at once to build context faster. Improved coordination across multiple tools and information sources enables the model to effectively leverage a wide range of capabilities in agentic search and coding workflows.
Advanced context management: Sonnet 4.5 maintains exceptional state tracking in external files, preserving goal-orientation across sessions. Combined with more effective context window usage and our new context management API features, the model optimally handles information across extended sessions to maintain coherence over time.
hi @harrison-halperin_LP several of the listed capabilities from the API are integrated in Cursor or are available to the model by default.
The Extended Autonomous operation may work and appear differently in Cursor vs on API or in Anthropics Claude.
1 Like
Claude Code’s extension for VS Code now offers a chat like Cursor and Copilot. I haven’t tested it myself, so I don’t know if it’s any good.
Hi, I not sure I meant Claude Code CLI. I meant Claude 4.5 Sonnet and starting with specs and small files is the key. Do not let your files grow in size. File size at max 100 lines to avoid choking. It is also more cost effective for you. Text replacement becomes difficult for agent when files size if big. In a small size, it will rewrite the whole thing fast. In a large file it tries to replace and it fails and fails. At least that is what I am seeing.
It will rush through the initial tasks too, instead of patching a file once per sub-task, it will sometimes just output the whole file in one go at the risk of a tool timeout and complete fail.
I had Sonnet 4 and Sonnet 4.5 do the same tasks. Then I asked them to review both their work and score the results. Sonnet 4.5 costs 30-50% more even though it sometimes forgets to do some parts or cuts corners aggressively.
They both cross-ranked Sonnet 4 at 100% and Sonnet 4.5 at 90%. They both praised Sonnet 4. They both don’t listen to rules and prompts and just barge through the plan head to tail without stops.
I love the Sonnet 4.5. It is super good. Much more better than the 4 was. It is difficult to point finger where and how, but it feels it gets better the idea and can now actually tell when something is not going to work (in my idea). Even less mistakes, but not completely free of them.
And I just tried the new Browser function. Oh my god how good it was! It really rock my world. Claude was completely able to navigate and even uploaded my test file needed in the service. Then it made the conclusions. Only “problem” was that it did not read debug window where it redirected debug information, but I supposed I could have asked another means for that if I knew.
1 Like
Running it now. Me likey. I blew $2.5k on Cursor last month. I need a more affordable maintenance tool now, this fits the bill.
2 Likes
THIS IS INSANE! Cursor is responsible for this! the API does not cost this much!
When you enter the casino with a hefty bag of cash and feel completely lost.
1 Like
Claude Sonnet 4.5 costs the same as Sonnet 4.
It’s pretty clear this is either due to poor caching or because caching isn’t being factored into the cost calculation.
I think it’s time for Cursor to comment on the situation. Is this difference in costs really expected model behavior?
I have continue testing it with cursor(I’m testing different solutions for my company). After 2 days of full work with 4.5 and cursor and normal task in flutter, typescript and python development I have spent the ultra plan +450 USD so far. So my estimation it will cost 5000USD+ / month using it.
It’s performance is great, but price VS work the gpt5 is probably 1/7 cheaper then sonnet 4.5 with cursor
1 Like
There are cache puts (forget the term) and cache read. Cache read is cheap, cache put is often more expensive than not using cache. So your base tokens can be higher because if cache use if cursor puts lots of context into cache and then it’s never reused.
Cache writes are more expensive but one time per input where cache reads are multiple times and this mostly affects high amount of cache reads.
Cache or no cache is shown in the request details in Usage.