Arsh Goyal (@arsh_goyal) - Twitter Profili | Zamantika Mersobahis Locabet

Sabitlenmiş Tweet

Arsh Goyal@arsh_goyal·6h

love this. i feel benchmarks have become marketing slides at this point, so letting people judge the real work themselves is new and this is doing that. will be interesting to see if the crowd verdict matches the official leaderboards and also curious how they'll handle voting bias at scale though. #HyperagentPartner

Hyperagent@hyperagentapp

Most benchmarks just give you a number on a chart. We made one where you judge the actual work side-by-side. Here's Fable 5 vs GPT-5.6 Sol on a mix of design, writing, and creative work. Try Hyperbench and choose your winner: hyperagent.com/s/rcW3h5uutU4B…

English

4

2

9

1.8K

Arsh Goyal retweetledi

Arsh Goyal@arsh_goyal·6h

love this. i feel benchmarks have become marketing slides at this point, so letting people judge the real work themselves is new and this is doing that. will be interesting to see if the crowd verdict matches the official leaderboards and also curious how they'll handle voting bias at scale though. #HyperagentPartner

Hyperagent@hyperagentapp

Most benchmarks just give you a number on a chart. We made one where you judge the actual work side-by-side. Here's Fable 5 vs GPT-5.6 Sol on a mix of design, writing, and creative work. Try Hyperbench and choose your winner: hyperagent.com/s/rcW3h5uutU4B…

English

4

2

9

1.8K

Arsh Goyal@arsh_goyal·6h

i feel benchmarks have become marketing slides at this point, so letting people judge the real work themselves is new and this is doing that. will be interesting to see if the crowd verdict matches the official leaderboards and also curious how they'll handle voting bias at scale though.

English

0

59

Hyperagent@hyperagentapp·7h

Most benchmarks just give you a number on a chart. We made one where you judge the actual work side-by-side. Here's Fable 5 vs GPT-5.6 Sol on a mix of design, writing, and creative work. Try Hyperbench and choose your winner: hyperagent.com/s/rcW3h5uutU4B…

English

14

6

38

28.5K

Arsh Goyal@arsh_goyal·2d

i am loving whats happening this month

GIF

Claude@claudeai

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

English

0

4

775

Arsh Goyal@arsh_goyal·2d

@vadiamit See you!

English

1

0

2

269

Amit Vadi@vadiamit·2d

yesterday set the bar. next up: I/O Connect Bengaluru. if you’re building, come say hi

English

19

1

81

4.3K

Arsh Goyal@arsh_goyal·2d

Brunch with @Tejasvi_Surya in Bangalore today and we had an extended chat around some very critical topics affecting all of us. These dialogues are important for giving a platform to talk about certain issues for people from all walks of life. We talked about: - AI in Political Systems - Job Losses and government policies for AI - Research Funding - Education Reforms - Upskilling and Reskilling and more. What is the one question you would ask your MP if given a chance?

English

1

0

11

1.2K

Arsh Goyal@arsh_goyal·3d

ok the Meta Iris thing is actually wild, a social media company shipping a new AI chip every 6 months through 2027. google proved this playbook with TPUs. inference doesn't need CUDA and inference is where meta's money burns.

English

1

0

3

628

Arsh Goyal@arsh_goyal·3d

@yunhaojiao crazy

English

0

151

Yunhao Jiao@yunhaojiao·3d

didn't expect 44 shipped builds in 10 days, still catching up on saying thank you to everyone who showed up for this.

TestSprite@Test_Sprite

Season 03 is officially closed. 🏁 Submissions for the hackathon wrapped up today. The $5,000 prize pool is locked in, and the race for the winning builds is officially on. The final tests have been run, the last commits are shipped, and the loop is complete. Massive thank you to everyone who built with us!

English

4

2

11

1.6K

Arsh Goyal@arsh_goyal·3d

@ai_for_success 👀👀

QME

1

0

1

196

AshutoshShrivastava@ai_for_success·4d

Thank you, Google DeepMind team, for the Nano Banana 2 Lite stickers! 💙 #GiftFromGoogle

English

16

0

94

6K

Arsh Goyal@arsh_goyal·4d

@hayyantechtalks

GIF

QME

0

1

18

Hayyan@hayyantechtalks·4d

@arsh_goyal Hy3 model by Tencent looks promising and will try this weekend.

English

1

0

1

50

Arsh Goyal retweetledi

Arsh Goyal@arsh_goyal·4d

Spent the last two days testing Hy3, the new model Tencent just launched. 295B MoE but only 21B active params, 256K context, and it's free on OpenRouter till July 21. I used the preview back in April so I wanted to see what actually changed. Quick thread on what surprised me:

English

7

2

15

2.8K

Arsh Goyal@arsh_goyal·4d

@MindWisdomMoney

GIF

QME

0

24

Maxwell 🕊@MindWisdomMoney·4d

@arsh_goyal Amazing. Really like the look of the hy3 model by Tencent. Il certainly be spending time this weekend trying it out!

English

1

0

1

1.1K

Arsh Goyal@arsh_goyal·4d

@shushant_l must try

English

0

1

26

Shushant Lakhyani@shushant_l·4d

@arsh_goyal Hy3 model by Tencent looks super cool and will test it this weekend

English

1

0

1

370

Arsh Goyal@arsh_goyal·4d

It's free on OpenRouter for another few days if you want to run your own tests: openrouter.ai/tencent/hy3:fr… Would love to hear what you build with it. @TencentHunyuan #Hy3 #Hunyuan

English

1

0

5

212

Arsh Goyal@arsh_goyal·4d

Honest take: this is not the model I'd pick for huge repo level refactors, there are stronger options for that. But for frontend work, long docs, and agent workflows, the cost to performance ratio is hard to argue with. It has three reasoning modes too, so quick tasks stay fast.

English

1

0

5

196

Arsh Goyal@arsh_goyal·4d

The agent stuff is where it clicked for me. Gave it one prompt: plan a feature, break it into tasks, write the code, then summarise next steps. It did the whole loop without me babysitting each step. Tool calls stayed stable too, which is usually where these workflows fall apart.

English

1

0

4

46

Arsh Goyal@arsh_goyal·4d

Then the 256K context test. I dropped a 70 page spec into it and asked for a build plan with timelines and open questions. It pulled details from page 60 that I had forgotten were in there. Most models lose the plot around the middle of a doc this size. This one didn't.

English

1

0

5

244

Arsh Goyal@arsh_goyal·4d

First test: asked it to build a small analytics dashboard. React, charts, filters, the usual. The preview used to mess up interaction logic halfway through. This one got the state handling right on the first pass. Pasted a console error, it found the actual cause instead of guessing. That part genuinely surprised me.

English

1

0

4

121

Arsh Goyal@arsh_goyal·4d

@elonmusk @battleangelviv and now with you quoting it, the views would go 20M+ in a few hours

English

1

0

7

487

Elon Musk@elonmusk·4d

𝕏 is a great platform for product announcements, especially if done by the CEO directly. Way more interesting to the public than generic press releases. This post by Mark Zuckerberg already received over 12 million views for free!

Mark Zuckerberg@finkd

(1) Today we're releasing Muse Spark 1.1 -- a strong agentic and coding model at a very low price. It's available through our new Meta Model API and in Meta AI.

English

4.3K

4.8K

61.2K

6.2M

Arsh Goyal

Keşfet