🚨API IS CHEAPER THAN SUBSCRIPTION. I ONE-SHOT A PLAYABLE MARIO CLONE WITH ONLY $2 OF API CREDITS. Yesterday I posted that Claude MAX 20x gave me less than 4 hours of actual Fable 5.1 usage before I was already at ~77% of my weekly limit. I was genuinely asking... how “45% cheaper” mattered if I could barely use the model. MixRoute AI saw the post and reached out to let me stress-test their API for a full day. So I thought why dont we make a comparision between both of them Claude Code session: $10.25 cache hit 97% MixRoute AI API usage: $2.00 Claude Code: ~25 min → $10.25 MixRoute: ~32 min → $2.00 ~6 minutes slower. ~80% cheaper. Both built a playable Mario clone. Full run sped up ~60x ↓show more

is also running a 100% top-up bonus right now. yes you literally get 2x of whatever you pay for. $10 → $20 $100 → $200 $500 → $1000 New accounts get $5 free, and using my invite gives another $5:

this was the claude code session stats too:

That's crazy bro

i never imagined i could use FABLE 5.1 with 2$? like??

cool comparison. crazy how much cheaper the MixRoute API is for similar results. nice one!

yes. it was the caching of fable 5.1 that made it cheap. but the 100% bonus on their sign up made it like even 2x cheaper than it is. which makes the the overall like 3x~5x cheaper than subscription.

that's impressive how the caching helped cut costs so much. those bonuses really make a difference too

@AbdoKerdawy You can make use of them too if you sign up today.

@AbdoKerdawy the caching made it so much cheaper for you. that bonus on sign up really maximizes the savings too

@MixRoute_ai yes!!

API costs significantly lower than subscription models.

yes the caching thing made it actually much heaper wtf??

Such an insightful comparison. Paying $2.00 and getting this good output means the AI is getting cheaper. We just need to find a model and an API provider like @Mixroute_ai that can make this good output possible at such a lower cost.

@MixRoute_ai yes 100%, i believe now at this rate, 100% top-up bonus AND 45% cheaper caching on API than on the sub plan actually made fable 5.1 (the most expensive model) usable through api

$2 vs a weekly cap you hit in 4 hours is the whole argument, the api just makes the ceiling visible

i think, seeing this "45% cheaper caching on API" (in the ss down) and then seeing @MixRoute_ai's "100% top-up bonus" and then doing a playable mario with 2$ makes me reconsider Claude's max 20x plan tbh

@MixRoute_ai max feels infinite until the weekly bar turns red. $2 mario is the receipt that sticks

@MixRoute_ai yes exactly 100%

i would be happy to share the full transcipt and full video if you want! however trying it yourself wouldnt be bad too. try it and tell me what you think!!

Difference between subs and api is just to big. It is in order of getting hundred or even thousands times more the price of sub in api price. I was recalculating my outputs on a frw different ones and it is always the same. Doubt it can be made equal so easily.

Damn, @MixRoute_ai is really a game changer.

@MixRoute_ai 100% Try it, they already gave me (and everyone haha) 100% bonus on top up for the rest of September

$2 for a mario clone insane

Yes exactly??

right

The real comparison is cost per useful output, not subscription price. If the result is comparable and the API is ~80% cheaper, that’s a hard number to ignore.

well. it was the same fable 5.1 model. and the same input prompt. yet the "45% cheaper caching on api based" + @MixRoute_ai's double credits on every credit you top up made it clear that i could actually continue using chatgpt for example specially after astra, and keep all my fable work on api based since it is more reliable and trackable and moslty less expensive and literally has no "50% of your weekly limit" cap thing just give it a try from my invite link🤗it will give you 5$+5$ anyway so you could literally replicate my experiment!!!!

✅🔄

YESS<33

interesting 👀

yess!!!! give it a try and tell me!!! 5$ for free is already included, and another 5$ if you use my invite link^^

with the initial 5$ only some models are available & all are like pretty outdated like a year old gpt-4o? and gemini 2.5 flash is there? also ur invite link didn't really work it seems fast tho let me test with what is there

no? they literally have fable 5.1 and even gemini 3.8 that got released yesteday? i was using fable 5.1 in the video, are you sure you configured it correctly?

yes the models are there but not with the initial 5$ credit

It worked with me? I will ask them @MixRoute_ai You too you may also DM them

It’s cheap cause it uses chinese proxies underneath

it wasnt cheap.. it is typical fable pricing with API caching if you know about it. the tokens used in each justifies it. just try it for yourself and you will understand. its free to try anyway

Damn, 80% cheaper for just 6 minutes slower is actually wild.

yeaa. also i think it was "slower" because the fable on API just decided to think longer hahaha.

This is sick bro 🔥

thank youuuuu

use omp with Muse Spark 1.3 Contributor very good)

what is omp? also thank you for the advice. i tried spark 1.3 contributor and i actually made a comparision i am finalizing now and i will post about it too, however, fable is still fable in the end. thats what i always end up feeling after seeing every "fable-level model" lol.

oh my pi

Hang on --- can you unpack the compare? Did you normalize your usage % into the sub amount? ie $200/mo = $50/week. If prompt A consumes 50% of weekly usage that's $25. Then compare same prompt to router and it's $X? If so: Great way to test and also - Wow!

Ofc!! Ill do the math and tell you yes!

Crazy that an API can be *less* than the sub. You should do a few more of these. Would be interesting to have more than an n=1

Depends on TTL, afaik from Claude code defaults to 1h, but for agentic heavy sessions 5m ttl is much cheaper. All you have to do if on Subscription is use 5m ttl + a keep alive probe

@grok wtf is this ? Verify and please tell me how

Claude pricing is insane, i switched to gpt 5.6 and works well for me

Yes exactly, and now without astra they give a banked reset everyday!!!

I get 100k USD worth of API usage on a 200/mo plan. What about you?

if this "100k usd worth of api" from ur 200$ plan then i would be happy to buy ur acc. my acc gives me barely few hours literally. and im on 200$ too

Skill issue is my first guess. Selling accounts violates TOS. I get ~40B tokens a month. Then maybe some of it is cached. They don't reveal the $$ amount API equivalent.

Bro the selling part was joke obv<3

It’s probably the harness

Well the harness was claude code my friend

Great man!

Thank you!!

the 97% cache hit is doing a lot of work here—were both runs given identical prompts and tool calls, or did the cheaper API path simply need fewer iterations?

Yes. Same prompt. Wait i will send you in another comment why this happened.

can you use the claude code harness with the api

Yes, i saw ppl do it with claude code cli withchanging some environment variables, i didnt try it but 100% it works

我今天尝试用它给我做几个游戏特效！很令人惊喜，它一次就能实现我想要的东西，代价就是，我现在cursor额外模型的额度已经没了哈哈哈哈

Yeaa hahahahaha ALSO WHEN DID YOU GET THE VERIFIED BADGE??? CONGRATZZZZ

今天想着整个蓝V好了，以后多和你分享AI开发游戏内容！

I AM SO HAPPY FOR THIS HAHAHAH. CONGRATZ AGAIN!!!

Thank you for giving MixRoute a try. We see you liked it very much. 🔥

I really did!! thank you for letting me try it, and im looking forward to top up ALOOTTT with this 100% bonus!!

We are happy to hear that. Real world usage is the best proof that a product really works. 📈

smells fishy

your post reads like you had a stroke while writing it

this is so good 🙌

is it because of cache hits or how is this even possible? @grok can you explain?