Nick Trenkler
opus 4.7 launched. thehype reviewed early x feedback. here’s what users are actually seeing:
1/ consumes ~50% more tokens than opus 4.6 and ~2x vs gemini
efficiency is getting worse, not better. cost per task is increasing with capability
2/ hits 5-hour limit fast
usage caps become a real constraint. even paid tiers can’t sustain heavy workflows
3/ weaker or uncertain long-context performance
4.6 positioned as a stable long-context model
4.7 introduces doubt in one of its core strengths
4/ still far from mythos-level autonomy
opus 4.7 can modify code, but not maintain full system integrity. engineer supervision remains required
5/ no control over adaptive thinking
reasoning depth is managed by the system. users lose direct control over cost and behavior during tasks
6/ issues with web search activation
no comments
7/ strong at async work and instruction following
better consistency across longer tasks. clear improvement toward agent-style execution
p.s. author of the tweet works at anthropic =)
8/ generates high-quality design and animations
output quality in visuals and frontend noticeably improved. models are becoming usable for presentation-level work
9/ can handle complex environments (e.g. minecraft), but unreliably
capability is there, but execution is inconsistent. still not production-grade autonomy
10/ excels at simple game building
strong performance on contained systems (shooters, simulators). works best where scope is limited and feedback loops are tight
what’s your take on opus 4.7 so far?
