Hasin Hayder

Founder at TyroLabs, TyroPlugins & HappyMonster. Built Tyro-Dashboard, Tyro-Login, Tyro-RBAC & Tyro-Checkpoint for Laravel. Creator of TyroPress.

[Hasin Hayder @hasin]

@opencode DeepSeek V4 Pro is still being charged at the old rate. Please update the pricing. @opencode

[Hasin Hayder @hasin]

@just_unmute Is there any update on the PR merge?

[Hasin Hayder @hasin]

@DeepTechTR Do you even read your own post? Just check how absurd you sound here.

Just because Pewdiepie released it, it does not need to be exaggerated

[Hasin Hayder @hasin]

Here's a sample request bill from CommandCode using DeepSeek V4 Pro just two days ago.

Input: 75,045 tokens
Cache Read: 74,368 tokens
Output: 3,000 tokens
Total Charged: $0.0032

Since the reported input already includes cache-read tokens, the actual cache-miss input is:
75,045 - 74,368 = 677 tokens

Applying the current DeepSeek V4 Pro pricing:

Input: 677 × $0.435/M = $0.000294

Cache Read: 74,368 × $0.003625/M = $0.000269

Output: 3,000 × $0.87/M = $0.002610

Total:
$0.000294 + $0.000269 + $0.002610 = $0.003173
Rounded to 4 decimal places: $0.0032

So unless I'm missing something, this invoice reconstructs almost perfectly using DeepSeek's current published rates. The claim that requests are being billed below the current DeepSeek V4 Pro rate doesn't seem to match the numbers. The usage breakdown itself shows you're charging at the model's standard rate.

And for $1, users are getting upto $10 usage of DS4, not $40.

I am not wrong, sir. Your product is genuinely good and provides real value to developers. I have no disagreement there. My concern is with the marketing. The way DS4 and Mimo usage is being presented gives users an impression that doesn't match the current pricing reality.

Peace.

[Hasin Hayder @hasin]

@skeletor_ong @CommandCodeAI No, it doesn't. If you want proof, check your usage dashboard. Charged at the current DS4 pricing. Your $1 effectively stretches to about $10 of usage, and you can consume up to $10. That's it.

CommandCode is great, but the marketing around DS4 and Mimo is misleading.

[Hasin Hayder @hasin]

Yes, when the pricing was higher, that statement was accurate. But the provider has since permanently reduced the price, so it no longer reflects reality. Continuing to advertise the old value gives users the impression they're receiving far more usage than they actually are at today's rates.

[Hasin Hayder @hasin]

@vitaliidodonov MRR

[Hasin Hayder @hasin]

@ryancarson I’ve been using Gemini 3.5 Flash for real-world work. If your prompt is good, your agents.md file is well-written, and your instructions are clear, trust me, it produces fantastic output.

[Hasin Hayder @hasin]

@just_unmute Just sent a PR that fixes the tray icon issue: https://t.co/iYtlxoK1Cq

Now Open unmute from the tray menu properly restores the window. Mo extra overhead.

[Hasin Hayder @hasin]

@JonBuildsHQ Codex, hands down!

[Hasin Hayder @hasin]

@just_unmute Great work, by the way. I've been a Flow user for a long time and am planning to switch. My only wish is that it supported Bangla.

[Hasin Hayder @hasin]

@munim I suggest giving Qwen 3.7 Max a try. It’s a genuinely high-quality model that can hold its own alongside Sonnet and Opus.

[Hasin Hayder @hasin]

Does anyone else feel like the new MiniMax M3 Token Plan pricing is a huge disappointment?

[Hasin Hayder @hasin]

@munim Are you referring to those benchmarks? Sadly It's not on par with Sonnet (actually far from it)

Even GLM 5.1 produces better result than M3

[Hasin Hayder @hasin]

@jayair But you guys seriously need to improve the support. We mailed a billing issue - it's been almost 10 days, no one replied.

[Hasin Hayder @hasin]

@AndrewK404 @bridgemindai Monthly token limit is also one third, LOL

[Hasin Hayder @hasin]

এআই নিয়ে কাজ করতে হলে, মডেলের পাশাপাশি হারনেসও কিন্তু অনেক ইম্পর্ট্যান্ট।

আমরা বেশিরভাগ সময়েই শুধু মডেল নিয়ে কথা বলি। কিন্তু মডেল আসলে পুরো ইকুয়েশনের শুধু একটা অংশ।

আসল পার্থক্য বেশিরভাগ সময়েই আসে ভালো একটা হারনেস থেকে
- প্রম্পটিং
- কনটেক্সট ইঞ্জিনিয়ারিং
- টুল অর্কেস্ট্রেশন
- মেমোরি
- ইভ্যালুয়েশন
- ওয়ার্কফ্লো / এজেন্ট ডিজাইন

একটা বেটার মডেল তো অবশ্যই হেল্প করে। কিন্তু একটা বেটার হারনেস সেই মডেল থেকে বেস্ট আউটপুট বের করে আনার জন্য একদম মাস্ট। আপনি খেয়াল করলে দেখবেন যে সেম মডেল উইথ সেম প্রম্প্ট একটা হারনেস এ একরকম আউটপুট দিচ্ছে, আরেক হারনেস এ আরেকরকম।

সো বেস্ট এআই প্রোডাক্টস আসলে শুধু কোন মডেল ইউজ করা হচ্ছে তার উপরেই ডিপেন্ড করে না, বরং মডেলের চারপাশে কী সিস্টেম বিল্ড করা হয়েছে, এবং সেগুলো দিয়ে কিভাবে মডেল এক্সেস করার হচ্ছে তার উপরও অনেকখানি ডিপেন্ড করে।