Post

Varun Mohan

@_mohansolo

Jun 3 • 1 tweets • 1 min read • Read on X

With less than five days of notice, Anthropic decided to cut off nearly all of our first-party capacity to all Claude 3.x models. Given the short notice, we may see some short-term Claude 3.x model availability issues as we have very quickly ramped up capacity on other inference providers, but we believe we have now secured sufficient near term capacity. We have been very clear to Anthropic that this is not our desire - we wanted to pay them for the full capacity. We are disappointed by this decision and short notice.

Gemini 2.5 Pro (now very high quality on Windsurf, new 0.75x promo rate), GPT 4.1, and more are all unaffected. We look forward to our continued partnership with all model providers.

• • •

Missing some Tweet in this thread? You can try to force a refresh

This Thread may be Removed Anytime!

Twitter may remove this content at anytime! Save it as PDF for later use!

More from @_mohansolo

Varun Mohan

@_mohansolo

Mar 12

There’s been a lot of talk recently about how Windsurf’s context retrieval is better than other products. One rebuttal I’ve seen is that all products “index your codebase”.

But indexing code ≠ context retrieval. It is necessary but not sufficient.

Thought I’d share a bit about what we’re doing under the hood to get the best results.

Indexing & embedding search is a tablestakes RAG technique. Btw, even for this technique there are approaches that make this more or less effective. One thing we are doing is AST parsing code and chunking along semantically meaningful boundaries - not random blocks of code. This means that when a code chunk is retrieved, it is a full function or class, not just an arbitrary block of consecutive code.

But embedding search becomes unreliable as a retrieval heuristic as the size of the codebase grows. Instead, we must rely on a combination of techniques like grep/file search, knowledge graph based retrieval, and more. With all these heuristics, a re-ranking step also becomes needed where the retrieved context is ranked in order of relevance. We use LLM based reranking under the hood.

Read 5 tweets

Share this page!

Enter URL or ID to Unroll

Varun Mohan

Try unrolling a thread yourself!

More from @_mohansolo

Varun Mohan

Did Thread Reader help you today?

Don't want to be a Premium member but still want to support us?

Send Email!