@codingdodobird M3 울트라 96기가 4테라짜리는 1150만원입니다
Spark 128기가 1테라짜리는 850만원입니다
순수 bf16 자체는 맥 스튜디오가 빨라도 NVFP4에서의 속도 + connectX 확정성 생각하면 맥 스튜디오 구매가 쉽지 않습니다
심지어 애플은 M5 와서야 MLX를 통한 8, 4비트 연산 지원을 시작해서요…
Physical AI evals are starting to become a real category.
Until recently, most VLA evaluation was basically:
LIBERO → success rate → leaderboard.
Now we're seeing a much broader stack emerge:
• Allen AI — unified VLA eval across 18+ simulation benchmarks
• LeRobot — one eval interface across multiple sim benchmarks
• PhAIL — real robots + production metrics like throughput and failures
• Robocurve — independent, real-world robot evaluation
• RoboDojo — bringing sim + real-world evaluation together
The interesting part isn't just more benchmarks.
It's the move from:
“Can the robot complete this task?”
to:
“How reliable, fast and general is this system in the physical world?”
I think independent physical AI evals are going to become increasingly important as robot models start looking more and more similar on demos.
God. Reset is only on the 20th. 3 more days.
X and Reddit are boiling. This isn't funny anymore. Thousands of devs, paying hundreds of dollars, sitting at 0% with no answers.
We're not asking for charity - we're asking for transparency. @thsottiaux what's happening?
Insane pricing loop: Get $500 worth of AI tier for just $99 total.
SuperGrok Heavy ($300) + Cursor Ultra ($200)
1. Subscribe to standard SuperGrok ($30)
2. Open promo link while logged in ->
upgrade to Heavy ($69)
3. Claim Cursor Ultra bundle
Plug your Grok credentials straight into OpenCodex to run Grok Heavy inside Codex without subscription caps.
@thsottiaux Any plans to add or improve Computer Use and Appshots support on Linux and Windows, as well as orchestration where a larger model coordinates smaller subagents?
@AbdoKerdawy@thsottiaux There are many reasons why the app is better than the CLI—especially when you need to view images in a chat session or use computer use.