Here is my Claude Pro review...
retained text
Bought Claude Pro on Sep 29, 2026. 4 weeks done. Vibecoder review (retired DevOps, so I know enough how dangerous it can be). Used Opus as Orchestrator and DeepSeek in my OpenCode Go subscription as a workhorse. **TLDR: 4 weeks of Claude Pro ($20) = 4.24B tokens, $1,336 in API value. 67x what I paid. Opus 5.5 ate 90% of it. Half the bill is just re-reading my own repo from cache.** [Vorssaint app shows the stats in fancy way](https://preview.redd.it/i7xz7zbuszxh1.png?width=554&format=png&auto=webp&s=256dafa65c2ab5a22f7b34615a137d4c5e658873) Context: * Building a small e-commerce site in OpenCode with Muse Spark. Tried a friend's sub back on Opus 5 — hated it, hallucinated garbage. * Opus 5.5 dropped days later and it was a different animal so I paid up. * Week 1 I ran Opus-only on high/xhigh like an idiot. Never touched the 5hr limit. Weekly cap is the only enemy. |Week|Tokens|Out|Hit|API $| |:-|:-|:-|:-|:-| |1: Sep 29–Oct 5|1.17B|3.14M|98.5%|$414.61| |2: Oct 6–12|874M|2.62M|98.8%|$288.26| |3: Oct 13–19|1.07B|2.88M|98.9%|$346.67| |4: Oct 20–26|1.05B|3.33M|98.9%|$260.64| |Oct 27|69.6M|0.26M|98.8%|$24.78| |**29d total**|**4.24B**|**12.2M**|**98.8%**|**$1,336**| |Model|Turns|API $|Share| |:-|:-|:-|:-| |Opus 5.5|19,001|$1,210|90.6%| |Sonnet 5.5|3,854|$124|9.3%| |Haiku + Opus 5|192|\~$2|\~0%| [Claude stats](https://preview.redd.it/08s7z0vrszxh1.png?width=827&format=png&auto=webp&s=e5ede3233dea0711a77504b259cce8762141e98c) Bill split: cache reads $762 (57%), cache writes $353 (26%), output $220 (17%). Fresh input $0.17. Ehhh... **- Week 1: Unreal.** Opus found 80+ real bugs, deleted 10k+ lines of my spaghetti, added just a few hundred back. Burned $415 in ONE week — 20x the monthly sub. It was absolutely great week. **- Week 2: Dogsh\*t.** Same work, totally different experience. Opus somehow got dumber and hallucinated a lot. Usage got exhausted pretty fast. Two days spent staring at a 100% bar. **- Week 3: Copium week.** Ran Opus as orchestrator + DeepSeek V4.1 Flash as a workhorse. Limits stretched further, sure — but Claude still billed $347, my 2nd priciest week. Got so much quality work done. **- Week 4: Finally reasonable.** Cheapest week ($261). Replaced Opus with Sonnet and pointed DeepSeek V4.1 Flash as the workhorse again. Sonnet fumbles a lot, doesn't act proactively like Opus. xhigh effort is a scam (tried few times, never again). Sonnet High and Opus medium/high are the ways to go. Launched the website in the week. **Bottom line: $1,336 of compute for $20. Insane deal!. Weekly cap totally runs the show.** *Method: parsed \~/.claude/projects JSONL, deduped by message ID. Claude's own stats say 9.6B because it counts every content block separately — real is 4.2B (Not my words but AI says so). I used a bit of AI to rewrite my post as English isn't my native langauge and also I am not an AI expert.*
document fetched 2026-10-28 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1xgpmx3/here_is_my_claude_pro_review