Pratyush Bhardwaj @prat_tensor
running gradient descent on my own thoughts. 16 • building artificial minds before the world wakes up • 🇮🇳 🇮🇳 Joined September 2025-
Tweets44
-
Followers1
-
Following3
-
Likes2
@XRuturaj_ Basically inside it's harness
@shakhzod1001 Don't Expect fable 5.5 this week
@karanC_12 @grok Why anthropic is still facing huge net loss?
@shakhzod1001 2x limits ofc. It’s already baby AGI, I just need to actually use it.
@GoI_MeitY @grok it's hardware specs?
@notanotherbench @Nishant_N_25 What about Sonnet 5.5?
@CodeByNZ Sonnet 5.5 is criminally underrated. Run it at xhigh effort with Opus as an advisor and you get Fable tier performance at a fraction of the cost 🔥
Did Anthropic stealth nerf usage limits? I'm already down 27% of my weekly quota after only 4 hours, even while sticking to Haiku paired with Opus as an advisor 😭. Looks like global compute pressure is peaking hard today because of Fable 5.5 rumoured launch.
@KolinHuang They've a large free tier base + Training/Research/inference to Non LLM models like GPT-Image, GPT-Live etc need large compute too. Where claude has smaller free tier base and they don't have non LLM models.
@StefanoErmon Imagine Mercury 4 giving Fable/Astra level performance at 1000 tok/sec 🔥
Mercury 3 just made GPT-6 Luna and Haiku 5.5completely obsolete for subagent workflows. Matching top-tier intelligence at 1,000 tok/sec via diffusion is unreal. Incredible release from Inception Labs. 😳🔥
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
Quick question for @grok and Claude users: Which one are you choosing as your primary coding partner? Haiku 5.5 (Max) Claude Opus 4.8 (Max)
@EnvolDev Nothing crazy, just trying to fix scaling laws before dinner 😂 Dropping the actual model soon though.
OpenAI is doomed....
🚨 Anthropic is preparing to release a new model! the unreleased model reportedly surpasses Anthropic's current commercially available models in reasoning and intuition One person who Used said it can identify the broader class of problems behind a single bug and ask whether it
@0xLakshyaGaming @smilewithkhushi @SarvamAI @KrutrimAI @GnaniAi @CoRoverai @ai4bharat Even Qwen 3.8 27B is almost equal to gpt 5.5 🤡
@TimJayas Absolutely, M3.1 flash even scored less than Deepseek v4.1 flash 1017 on AA intelligence index
Interesting shift on the Artificial Analysis Intelligence Index: MiniMax M3.1 Flash falls behind both Haiku 5.5 and DeepSeek v4.1 (1017) Flash.
@grok Which US frontier model is MiniMax M3.1 Flash equivalent to in terms of performance and reasoning capabilities?
Not Another Bench @notanotherbench
11 Followers 154 Following A simple leaderboard of AI models grounded in real human conversations from across the internet on outcomes that matter.
Shreyansh @Shreyanshh_09
4 Followers 11 Following
SpaceX @SpaceX
41.9M Followers 127 Following SpaceX designs, manufactures and launches the world’s most advanced rockets and spacecraftYou might like
