Feels like Anthropic is trying to differentiate from OpenAI by building a small number of high-quality first-party things instead of trying to enable an ecosystem that may be harder to capture value from.
500k context is enormous, but at 3.5 Sonnet prices, also pretty expensive:
$1.50 for one query maxing that out. However, if you use their new prompt caching setup and cache most of the big prompt and are hitting it frequently enough, it's more like 15 cents per query. That's still a lot, but it could produce very valuable results, and sticking the whole thing into the prompt instead of trying to narrow it down first with traditional search lets you skip all the pain of indexing and vector DBs and whatever. This makes prototyping much easier because you can get good results right away by jamming your whole corpus into the prompt, and only then decide whether optimization is worth it.
> The Claude Enterprise plan offers an expanded 500K context window, more usage capacity, and a native GitHub integration so you can work on entire codebases with Claude.
Woah. This is huge! I thought their biggest models only offered 200k context windows?
Is this the first time we've seen an LLM provider put their newest/best models behind a SaaS wall before they make it available through API?
Same for me with Firefox and macOS. The form is getting blocked by Firefox's "Enhanced Tracking Protection" feature. This is the request that's blocked: "GET https://js.hsforms.net/forms/embed/v2.js"
Yeah, there's definitely a lack of perceived value with ChatGPT now, especially when you experience the difference between Anthropic's offerings, or try one of the open models that don't refuse every third prompt. Claude is the best value for your money, and definitely the most technically capable.
I’d love to be able to use the GitHub integration with a personal account. Currently I use claudesync that works pretty well but I’m figuring it’ll stop working at some point
500k context is enormous, but at 3.5 Sonnet prices, also pretty expensive: $1.50 for one query maxing that out. However, if you use their new prompt caching setup and cache most of the big prompt and are hitting it frequently enough, it's more like 15 cents per query. That's still a lot, but it could produce very valuable results, and sticking the whole thing into the prompt instead of trying to narrow it down first with traditional search lets you skip all the pain of indexing and vector DBs and whatever. This makes prototyping much easier because you can get good results right away by jamming your whole corpus into the prompt, and only then decide whether optimization is worth it.