← sources

Got my most expensive coding AI Sub i ever gotten after trying Alternatives

source: reddit (reddit:ClaudeAI)published 2026-09-29 15:23ingested 2026-09-30 02:38

retained text

[My new Subscription](https://preview.redd.it/h2crzo4a5hsh1.png?width=788&format=png&auto=webp&s=45862643742dbf33be7c525b826f0975a0e4e093)

So after using this professionally before, but at company cost, I finally pulled the plug and got myself a Max plan. I hope my wife will not castrate me with a spoon.

Here's everything I tried already (best to worst):

**Ollama Cloud ($20, old grandfathered subscription)**

[Ollama Cloud with Generous Quota Shown, the blue right side is flash model, i used pro model by mistake once and 300 requests cost a lot as they charge per gpu usage and pro needs a lot, so you can see that flash model 6.4k requests at around 150k context took almost nothing out of old subscription model.](https://preview.redd.it/xicycdb59hsh1.png?width=582&format=png&auto=webp&s=9c8db79562a95209e3e9025a123306e1970223e0)

Overall the best, cause of zero retention and blazing fast speeds. DeepSeek on it has been going at around 200 tps, which is amazingly fast when other platforms only allowed 80 tps and had issues. Mine has been running without any problems. It also has pretty amazing web search, I can use bigger models like Kimi K3 and I'm not bound to any one model. There's also a friendly Ollama command that auto-adds these to Codex, Claude etc. I'm considering keeping this sub.

Old sub 9.5/10, new subscription 4/10, since the prices went up and you don't get a good quota anymore, just hard limits. I tried it on another account and it was under 20% of the DeepSeek I got on the old sub. Praise though, you can use DeepSeek here without your data being sent to China since they self-host it.

**Opencode Go**

[Opencode Go 10$ sub](https://preview.redd.it/mwdclxky8hsh1.png?width=1023&format=png&auto=webp&s=b584ffbd6a858f8889c43f25808946f9e158568e)

Honestly a mixed bag. Amazing for the price, but because of the low price and the knife-edge balancing the devs are doing (which I'm grateful for), the limits were changing almost weekly. If you use DeepSeek and the cheapest models there you get far more than at other providers, basically $10 turns into $60, and you're not bound to anything. I'm considering keeping this sub as it's very cheap, has a lot of models from providers with very little downtime, and is overall transparent.

Until recently you could use it everywhere with their API key, but after the X-Headers got forced and they moved away from the OpenAI standard, it's limited in some cases with custom apps and needs a separate OpenAI-standard setup with headers for Opencode. I don't blame the devs, they already said this was because people were reselling the service, and that was a net loss on a service running on a knife-edge.

9.5/10

**CommandCode**

[Subscription Proof CommandCode](https://preview.redd.it/rr90nkg29hsh1.png?width=1746&format=png&auto=webp&s=33870adb03b8d112cb4f5f284f5569243bf422a9)

Better on paper than Opencode Go and others, but the portal is very weird and the calculations make no sense. Their $60 lasts around 20% as long as it does on Opencode Go, and the dev is known for making horrid comments to people and banning people outright. After trying the GOAT plan I burned through the limits with cheap models in around 10% of the time it took on Opencode Go, even though on paper it had more limits. I messaged the dev and got a reply that I didn't account for "cache", which I did. And even then cache is such a small part it's irrelevant. Their harness is also horribly benchmaxed and almost unusable.

1/10

**Codex via Plus sub**

[Codex Sub](https://preview.redd.it/y8zo0yhk9hsh1.png?width=858&format=png&auto=webp&s=77a0d9ba70a2ba6a2b21b2ee738eb7cafe7aae38)

A good subscription, with a big IF: you need an unlimited chatbot with amazing voice recognition, which it's currently close to (although that may change). Codex has always been very generous, but since the GPT 5.6 era I feel like the limits got cut by around 70%, and it's currently only usable with Luna. If you compare the $20 sub from Anthropic to the $20 Codex one, with Opus 5.5 you get around 4-6x the work done compared to GPT-6 Sol (Astra burns so quickly you won't manage to finish a smaller task).

The web search in it is honestly one of the best I've seen, it scrapes the web amazingly and I see nothing coming close. Only downside is the kinda bad terminal usage on Windows, on Linux it works great.

Grok improved a LOT since last few years and it has become something between Sonnet 5.5 and GPT 6-Sol in Terms of quality

**Clinepass**

[Clinepass Subscription Quota Panel](https://preview.redd.it/o0tzj4q99hsh1.png?width=1577&format=png&auto=webp&s=30696b9148b0daa6c1f128c29e772d2d9a949b43)

For $5 for the first month it's a steal. For $10 I'd rather go Opencode Go, since this is basically Opencode Go with around 40-50% of the quota, but still a steal.

8/10 after the first month, and 10/10 for the first month.

**Cursor $20 sub via yearly plan**

[Cursor Sub Panel](https://preview.redd.it/sr5a5rid9hsh1.png?width=1107&format=png&auto=webp&s=8223404e4a475b1906c2c7fe67cb91cd06cefaa4)

[Cursor Tokens used](https://preview.redd.it/rxwhgy3g9hsh1.png?width=1140&format=png&auto=webp&s=38ea83244500879cda4e6de39e10d85f82bc8f26)

This is a hidden gem. It gives you around $1700 of Grok 4.6 quota and $20 for other models. It's pretty great but limited to its own harness, which is honestly kinda bad, but the desktop app is a lot better and manages to hold its own. Currently it lets you use the Grok bot as a separate quota, which makes it around $4000 of Grok usage combined, and it's honestly amazing in speed, reliability and quality.

I got it via the yearly sub and paid via Klarna, so it came out to like $14 a month, which is a steal (Klarna had a 0% financing deal and I figured I'd use it for a year anyway). Since I was a student back then it was even better.

10/10

Antigravity:

"honestly this is so bad i do not want to even download antigravity again to do screenshots"

[Screenshot via Web to prove i got Pro, i actually got 3 Accounts from Family who got it via mobile provider but the model is literally so bad it is unusable and cannot be trusted at all.](https://preview.redd.it/zakftu6eahsh1.png?width=839&format=png&auto=webp&s=204d763d39b5f5ad52385cf785649c7cfcca3b9b)

So this is what i got cause i got it for free via my university for a year the 27$ sub then got it via mobile provider for 10$ for another year.

Gemini 3.1 Pro was pretty amazing but it has gone downhill gotr scaled way down and the model became insanely dumb like quantized, the flash models are high as kite and unusable honestly i would not recommend it for anything more than simple chat, only thing i can recommend is 8$ sub for AI Voice while Driving to learn about new topics like DHCP and Sysadmin stuff or news but for other stuff it is unusable. Gemini 3.8 Flash is an improvement but it behaves like a 9B model, do not believe the benchmarks.

3/10

I also tried subs that are no longer online, like the Alibaba Cloud $20 sub (basically unlimited Qwen, something ridiculous like 40k requests every 5h, I don't remember exactly, but it was godly, and it got discontinued) and the old MiniMax plans with 5k requests per 5h.

Now, after using the $20 Anthropic sub (coming back after the early Sonnet days, before Haiku), I must say this is around 6x the current Codex quota, with much better quality and way more organized, so much that I got the $100 sub. We also got Sonnet 5.5 right now, so this doubles it, making it the best deal currently on the market if you use it as an orchestrator for OSS models via Ollama Cloud, self-hosted Qwen 27B or Opencode Go.

document fetched 2026-09-30 00:22 · canonical https://reddit.com/r/ClaudeAI/comments/1wtd1tn/got_my_most_expensive_coding_ai_sub_i_ever_gotten