MakeHubAI

The Best AI Inference. Zero Compromise. Period.

[RomainB @Rom1Bat]

Yo I have seen people complain that we couldn't have unlimited strong model on chatGPT anymore

Simple fix. It seems to work. Not always a thinking model, but always a slow inference, so it seems like a decent solution

[MakeHubAI @MakeHubAI]

@Rom1Bat @OpenAI very small from you if true :(

[RomainB @Rom1Bat]

Isn't slopenai a great nickname for @openai?

Im trademarking it

[MakeHubAI @MakeHubAI]

We launched on Product Hunt!

https://t.co/tTMIz1RvL1: LLM Provider arbitrage to get the best performance for the $

https://t.co/8A2aHCqbIW via @producthunt

[MakeHubAI @MakeHubAI]

@Rom1Bat I am scared of that video

[RomainB @Rom1Bat]

https://t.co/tsHj4SqWSG

[RomainB @Rom1Bat]

🧐

[MakeHubAI @MakeHubAI]

@Rom1Bat that slider is pretty cool when you know what's behind

[MakeHubAI @MakeHubAI]

we do magic

[RomainB @Rom1Bat]

Introducing @MakeHubAI
You pick the model on any app and we make sure you get the best speed and latency for the dollar for the $.

We literally create a real time competition between model providers to deliver your inference. Feedback hyper appreciated!

[MakeHubAI @MakeHubAI]

@Rom1Bat wait does that mean I lose my job?

[Mistral AI @MistralAI]

We're proud to announce Mistral Compute—an unprecedented AI infrastructure undertaking in Europe, and a strategic initiative that will ensure that all nation states, enterprises, and research labs globally remain at the forefront of AI innovation.

Read more in the thread.

[antonbuild @anton_build]

most devs using ai tools still burn money & lose the few hairs they have left fighting to fix bugs; because they don't optimize their coding tools

I spent thousands in inference cost in last few weeks and came up with some tactics

here are the tips I share to my friends to use ai to ship code 3x faster, cheaper, smarter:

1. big model for planning, cheap model for execution: big ai models (Gemini 2.5 pro) are powerful planners. use them to outline & strategize tasks. then, delegate small, actionable tasks to cheaper models (Gemini flash 2.5).

2. small tasks, heavy upfront context: always feed ai short, specific tasks with detailed context upfront. this avoids ai wasting tokens (& your money) trying to learn your entire codebase each time.
pro tip: use voice-to-text (wispr flow) to deliver context fast. your speaking throughput is ~3x typing speed, incentivizing you to share richer context.

3. know your codebase, never blindly approve edits: ai makes mistakes. always read, verify, and understand what it proposes. blind approval leads to bugs & tech debt. stay critical, ship safely.

4. optimize servers with makehub: stop guessing infrastructure costs. use makehub to quickly find servers with the best tokens per second-to-cost ratio. this alone saves hundreds monthly.

5. centralize documentation, automate updates: place all your README files centrally, then provide them as context to ai. configure ai to update your README at task completion automatically-keeps docs fresh & coherent.

6. summarize aggressively: regularly request ai to summarize conversations (/smol or similar tools). it forces clarity, efficiency, and context coherence.
and finally:
cline beats cursor every time-here's why: cursor reduces your context to cut costs (compressing input tokens). cline aligns incentives better: richer context, clearer understanding, fewer bugs, more speed.

use ai intentionally, watch productivity skyrocket.

[RomainB @Rom1Bat]

This will accentuate in the next 6 months.

Only Anthropic will continue to use their own model to code. Everyone else will be on cheaper alternatives delivering the same value, ~2-3x faster and >10x cheaper.

[RomainB @Rom1Bat]

When using classical AI inference endpoint, you can get rate limited, throttled down, or even simply not get the best deal on the market.

With makehub{dot}ai you always get the best value for your money. We find the cheapest provider out-there for the speed and latency you want.

[MakeHubAI @MakeHubAI]

@Rom1Bat And it takes 20 seconds to setup!

All your agents will virtually get cheaper and faster on https://t.co/i13L8cFuan

[MakeHubAI @MakeHubAI]

@Rom1Bat @AnthropicAI we are kind of fast at this, but you still have to integrate with AWS and vertex :)

[RomainB @Rom1Bat]

@AnthropicAI you can try Opus and Sonnet already on https://t.co/V2EyU2Tzxi !

[MakeHubAI @MakeHubAI]

The new Gemini pro 2.5 is available on https://t.co/i13L8cFuan. It's better in coding, and slightly worst in everything else. Worth it if you are using it on @cline or Roo Code

[MakeHubAI @MakeHubAI]

@Alibaba_Qwen test it on https://t.co/i13L8cG1ZV!

[MakeHubAI @MakeHubAI]

Qwen3 already available on https://t.co/bpN55CpQ67!