James Swift

Anthropic just shipped Opus 4.7 today and i had to write about it somewhere because the jump is weird.

I ran the same backlog task on 4.6 and 4.7 back to back. same repo, same prompt, same tools. 4.6 looped on a bug for 25 minutes and was not going to solve it. 4.7 closed it in eleven, and the part that freaked me out is that it paused in the middle to sanity-check an assumption i had not asked it to check. literally wrote "before i write this migration, let me verify the actual shape of the response object, because my assumption here might be wrong" and then went and verified it. unprompted.

That self-verification behavior is the thing. Vercel is reporting it does proofs on systems code before starting work. Hex says it flags missing data instead of making up plausible-but-wrong fallbacks. Genspark measured loop rates on hard queries and 4.7 basically stopped looping. different teams, different harnesses, same pattern.

the numbers are nuts too:

I've used SocialBee, Buffer, Jasper, Copy.ai, and Castmagic. They all promise "your tone of voice". Most are just a ChatGPT wrapper bolted on to create quick captions, but none of them let you define specific rules and enforce them on every output. I'd spend 15-20 minutes per piece correcting the same mistakes the tool kept making.

SplitPost (splitpost.io) takes one piece of your content and generates platform-native posts for LinkedIn, Twitter/X, Instagram, Threads, Reddit, TikTok and Facebook, each in the relevant voice and formatting for that platform. You set voice rules (banned phrases, sentence pattern constraints, register checks) and every output gets validated against them before you see it.

Free tier, no card required. This is app three in 18 months as a non-technical solo founder in the UK. I'd genuinely like to know if the voice enforcement angle lands for anyone else or if I've built for a problem only I had.

Here s what changed:

Production-ready code with minimal oversight, and it can verify its own outputs
More control over reasoning effort
3x better vision (now up to 3.75MP images)
Improved instruction following and overall reliability
New xhigh reasoning mode for finer control between speed and depth

Same pricing as Opus 4.6 ($5 and $25 per million input and output tokens). The new tokenizer can use around 1.0 to 1.35x more tokens depending on content, though this can be managed through effort settings and task budgets.