Satvik Paramkusham

Teaching Gen AI at @BuildFastWithAI |

Building @unrot_co

[Satvik Paramkusham @satvikps]

🤯 I created an open-source implementation of @notebooklm!

🤖 Deepseek-V3 API using @openrouter
🎙️ @play_ht TTS using @FAL API
📚 Create AI podcasts on ANY topic
🆓 100% Customizable

All this in <50 lines of code!

⭐️ Check out the GitHub repo: https://t.co/Jp96O5aBci

[Satvik Paramkusham @satvikps]

🔥 Anthropic may be the fastest company to get near a $1T valuation.

But something doesn’t add up:

> Google Valuation/Revenue - $4.56T / $402.8B
> Meta Valuation/Revenue - $1.61T / $200.97B
> Anthropic Valuation/Revenue - $965B / $30B run-rate

> Anthropic is valued at ~32x revenue.
> Google is at ~11x.
> Meta is at ~8x.

Is this AI Bubble?

[Satvik Paramkusham @satvikps]

How many Gemini Omni videos can you generate on a $20 plan?

I started with a fresh account - no prior usage.

One video consumed 33% of 5-hour window limits. That's about 3 videos every 5 hours.

It also consumed 2% of weekly limit - which is ~50 videos per week! That's far better than earlier 3 Veo 3.1 videos per day.

If you were you only using Pro plan for generating videos:

$20 ~ 200 Omni Videos.

$0.1 for 10 second high quality and intelligent video models. It's a steal!

[Min Choi @minchoi]

Gemini Omni is insanely impressive.

People are unlocking new creative ways to use it.

10 wild examples:

[Satvik Paramkusham @satvikps]

@rahilnjain Exactly! Many treat it as filler before the real answer but it's actually a gold mine

[Satvik Paramkusham @satvikps]

Love the fact that Anthropic is displaying full reasoning traces for Claude Opus 4.8!

It helps me in two big ways:

> I now have something to read rather than waiting for the response or switching to a different window to run another agent. Makes it easy to stay focused on one task at a time.

and the next one is super useful

> Reading the traces helps me understand the gaps in my prompt and quickly identify if the agent is going in the wrong direction. This reduces a lot of back and forth with the model.

[Satvik Paramkusham @satvikps]

@lennysan Build Fast with AI

[Satvik Paramkusham @satvikps]

Options for your monthly AI Budget (100M tokens):

> GPT-5.5: $3000
> Claude Opus 4.7: $2500
> Gemini 3.5 Flash: $900

And the Chinese options:

> DeepSeek V4 Pro: $87
> MiMo-V2.5: $87

DeepSeek recently dropped the price by 75% permanently. 🤯 And now Xiaomi dropped the price by a freaking 99%!

Thats 35x cheaper than GPT-5.5 and 17x cheaper than Sonnet!

Looks like Chinese labs are flushing market with ultra cheap AI before OpenAI and Anthropic IPOs.

[Xiaomi MiMo @XiaomiMiMo]

🚀 Better inference efficiency, lower costs, broader access.

MiMo-V2.5 Series API pricing is now permanently reduced — by up to 99% compared to previous pricing.
✨ Unified pricing across all context lengths.
MiMo Token Plans have also been upgraded:
• 5–8× more usable tokens at the same price
• Simpler and more transparent billing rules
🎁 As a thank-you to current users, all current Token Plan credits will be fully reset.
🎧 MiMo-V2.5-TTS remains free for a limited time.
⏰ Effective May 26 at 6:00 PM PDT.

These improvements are powered by continued inference optimization and serving efficiency upgrades across the MiMo stack.
🛠️ We’ll also publish a detailed technical blog on the inference optimizations later — stay tuned.

[Satvik Paramkusham @satvikps]

Last week, I conducted AI training for 2 Fortune 500 companies!

My brief: Use Copilot since all our employees have licenses.

While prepping for the session, I tried my Claude workflows and oh boy - every prompt failed in Copilot!

> Copilot failed to create a simple PPTX file
> Couldn't use connectors
> It couldn't even use context from the Copilot project files
> Feels like I've been ported back to GPT-3.5 times!

Now, get this - Copilot is the default AI tool for many corporate companies.

Execs read about AI tools and their 10x productivity promise, and are pushing their teams hard!

But 10x productivity is happening for the teams using Claude/Codex, not Copilot.

The Twitter world and the corporate world are miles apart, but they have the same goals.

[Satvik Paramkusham @satvikps]

@antigravity Is it same as Gemini CLI? If not, why have two CLI tools? It is so confusing.

[Satvik Paramkusham @satvikps]

@TheoryoftheDog It can't even be compared to Claude Code and Codex. It is far behind.

[Satvik Paramkusham @satvikps]

I wanted to make a post on 6 use cases to Antigravity 2.0!

Started testing with few prompts around web design, code refactoring and content creation.

And then I gave up.

Antigravity 2.0 + Gemini 3.5 Flash feels like it is 2 years behind Codex and Claude Code!

Negatives
> UI look likes a notepad
> Very weak in web design tasks
> Asking permissions too many times even for pip commands
>Wasn’t able to handle simple refactoring tasks

Positives
> Flash is cheap
> Extremely fast (1000+ tokens per sec) on Antigravity

Overall, Antigravity 2.0 feels crippled due to lack of the frontier model. Will try it again after Gemini 3.5 Pro release!

[Satvik Paramkusham @satvikps]

@stevencheng

[Satvik Paramkusham @satvikps]

@BReal_01 I noticed it uses quite a lot of tokens to produce average output.

[Satvik Paramkusham @satvikps]

@prashant_hq I agree. Not worth using it for this price point with an API - go for Kimi or GLM

[Satvik Paramkusham @satvikps]

@JasonDeLio Lol, I build UI directly with Opus and Kimi all the time - they handle it just fine without a babysitter tool upstream. Stitch and Claude Design are crutches for folks who can't design with a frontier model directly. That said, I Iove Stitch for end-to-end design tasks.

[Satvik Paramkusham @satvikps]

@marcherdina I thought they changed Gemini CLI to Antigravity CLI. It's so confusing.

[Satvik Paramkusham @satvikps]

@ProgenticEngine Fair on compute. But I think a permanent cut is a one-way door. Once you anchor the market at this price, you can't quietly walk it back. The moment DeepSeek hikes 3-4x, the same cost-driven crowd that came for the price leaves for the next cheap frontier model.

[Satvik Paramkusham @satvikps]

For two years I told every client the same thing: don't optimize for cost. Just use the flagship model and wait.

The price always falls. I was right for two years. This year I was wrong.

GPT flagship OUTPUT price, per million tokens:

> GPT-4 (Mar '23): $60
> GPT-4 Turbo (Nov '23): $30
> GPT-4o (May '24): $15
> GPT-5 (Aug '25): $10

A 6x collapse in 28 months. "Just wait, it gets cheaper" was the safest advice in tech.

Then 2026 flipped it:

> GPT-5.4 (Mar): $15
> GPT-5.5 (Apr): $30

The frontier's price has tripled in eight months, back to 2023 levels. OpenAI, the same lab that sold us "intelligence too cheap to meter," has quietly walked it back.

DeepSeek is the only one still holding that promise out loud: a 75% cut to a frontier model, made permanent, while everyone else hikes.

[DeepSeek @deepseek_ai]

We are making our discount permanent! 🎉

Enjoy building with DeepSeek-V4-Pro and bring your innovative ideas to life! 🚀

[Satvik Paramkusham @satvikps]

@praveensinghv @Replit @paulg This is the exact thing I'm seeing across the board. The "just build, cost will fall" assumption broke this year.

[Satvik Paramkusham @satvikps]

🚨 Google just launched Fitbit Air for $99.

Everyone's calling it a Whoop killer.

But the real story: it's an AI agent platform for your health.

31 data points, sleep, HRV, SpO2, heart rate, all exposed via the new Google Health API.

Your wearable just became a tool for your agent.

[download unrot app for latest AI news -> https://t.co/XEcpDZl800]