Joined April 2010
- Tweets 1,881
- Following 395
- Followers 250
- Likes 1,728
200 Photos and videos
Pinned Tweet
Claude Fable 5 is out and this might not stop here.
Open-source answered by shipping the entire stack.
Tools, models, and the builders pushing it forward. One bookmark, the whole map π
π οΈ TOOLS β the goated 5
β’ Hugging Face β the town square. 500k models, datasets, Spaces β huggingface.co
β’ Ollama β run any model locally in one command (MIT) β ollama.com
β’ llama.cpp β the engine under Ollama, LM Studio & Jan β github.com/ggml-org/llama.cpβ¦
β’ vLLM β production serving, the throughput king β github.com/vllm-project/vllm
β’ Open WebUI β self-hosted ChatGPT UI for local models β github.com/open-webui/open-wβ¦
π§ MODELS β 5 worth running (mid-2026)
β’ DeepSeek V4 β leaderboard king. 1M context, MIT. Agentic long-context.
β’ Kimi K2.6 β #1 open on Artificial Analysis. Agent swarms, long autonomous runs.
β’ GLM-5.1 β best all-round agentic coder. Cleanest MIT.
β’ Qwen3.6 β best small dense coder. Apache 2.0. Your license-clean workhorse.
β’ MiniMax M3 β newest. Frontier coding 1M ctx multimodal. 59% on SWE-Bench Pro.
β³ Laptop-local? Gemma 4 12B runs in 16GB RAM.
π€ BUILDERS β follow these 5 right now
β’ @danielhanchen β Unsloth. 10M downloads/mo. Fine-tune 2x faster, 80% less VRAM.
β’ @ggerganov β llama.cpp / ggml. The reason AI runs on your own hardware.
β’ @Teknium1 β @NousResearch Research. Fastest growing AI agent. Hermes models open datasets.
β’ @reach_vb β Hugging Face. The fastest feed for new open drops.
β’ @ClementDelangue β HF CEO. The institutional voice of open AI.
Special mentions to my favorite creators (@thdxr (@opencode), @0xSero (local.ai) and @neural_avb
on just awesome learning content.
Notice something? Four of the top five models are from Chinese labs.
The rest of the world needs serious catch up!
Save this one. π
#OpenSourceAI #LLM #LocalLLM #BuildInPublic
1
626
neekhil vatsa retweeted
Claude Fable 5 is out and this might not stop here.
Open-source answered by shipping the entire stack.
Tools, models, and the builders pushing it forward. One bookmark, the whole map π
π οΈ TOOLS β the goated 5
β’ Hugging Face β the town square. 500k models, datasets, Spaces β huggingface.co
β’ Ollama β run any model locally in one command (MIT) β ollama.com
β’ llama.cpp β the engine under Ollama, LM Studio & Jan β github.com/ggml-org/llama.cpβ¦
β’ vLLM β production serving, the throughput king β github.com/vllm-project/vllm
β’ Open WebUI β self-hosted ChatGPT UI for local models β github.com/open-webui/open-wβ¦
π§ MODELS β 5 worth running (mid-2026)
β’ DeepSeek V4 β leaderboard king. 1M context, MIT. Agentic long-context.
β’ Kimi K2.6 β #1 open on Artificial Analysis. Agent swarms, long autonomous runs.
β’ GLM-5.1 β best all-round agentic coder. Cleanest MIT.
β’ Qwen3.6 β best small dense coder. Apache 2.0. Your license-clean workhorse.
β’ MiniMax M3 β newest. Frontier coding 1M ctx multimodal. 59% on SWE-Bench Pro.
β³ Laptop-local? Gemma 4 12B runs in 16GB RAM.
π€ BUILDERS β follow these 5 right now
β’ @danielhanchen β Unsloth. 10M downloads/mo. Fine-tune 2x faster, 80% less VRAM.
β’ @ggerganov β llama.cpp / ggml. The reason AI runs on your own hardware.
β’ @Teknium1 β @NousResearch Research. Fastest growing AI agent. Hermes models open datasets.
β’ @reach_vb β Hugging Face. The fastest feed for new open drops.
β’ @ClementDelangue β HF CEO. The institutional voice of open AI.
Special mentions to my favorite creators (@thdxr (@opencode), @0xSero (local.ai) and @neural_avb
on just awesome learning content.
Notice something? Four of the top five models are from Chinese labs.
The rest of the world needs serious catch up!
Save this one. π
#OpenSourceAI #LLM #LocalLLM #BuildInPublic
1
626
Dont read this if you think a frontier AI Lab in India is a money problem. For everyone else, see the truth in data
DeepSeek trained a frontier model for $5.6M. That's the media-rights value of roughly half an IPL match
I found a way for Indian companies to start building foundation models. It's cricket and Bollywood. Hear me out.
We keep saying India can't afford to train frontier models. Compute's too expensive. GPUs are imported. Talent's in San Francisco.
Cope.
GPT-4 cost ~$78M to train. Gemini Ultra, ~$190M. Even the 2026 frontier class tops out around $200β500M a run.
The IPL media rights for 2023β27? $6.2 billion. For 410 cricket matches.
That one TV deal could train every frontier model ever shipped β several times over β with money left for a victory lap.
India's entire AI Mission β compute, datasets, skilling, startups, all of it β is βΉ10,372 crore over five years. ~$1.25B. One IPL cycle is nearly 5x the national AI budget.
The 2025 box office crossed βΉ13,395 crore. Bollywood alone hit a record βΉ5,504 crore β in a single year. More than the whole IndiaAI Mission, spent on motion pictures, annually.
We have the money. We've always had the money. We just point it at a cover drive and three-hour-interval samosas instead of an H100 cluster.
And here's the part that should hurt: we also have the data.
70 million concurrent viewers on a single T20 final. Decades of Hinglish commentary, vernacular dialogue across ten languages, the densest behavioral dataset on 1.4 billion people's attention ever assembled.
Sarvam got 4,096 H100s and a βΉ98 crore subsidy to build a 70B model. An IPL franchise's salary cap is βΉ120 crore. The purse for eleven cricketers outranks the money that built India's flagship LLM.
The objective function of Indian capital is entertainment. Not compute.
So here's the pitch: we don't need a new fund. We need the @BCCI and @DharmaMovies to realize they're sitting on the best-capitalized, best-distributed, best-instrumented AI lab in the country.
The GPU money was never missing.
It was funding the cover drive.
1
3
763
What $6.2B could buy β 1,107 DeepSeeks, 79 GPT-4s, 33 Gemini Ultras from one TV deal.
1
34
Now tell me everyone that a frontier AI lab is not a money problem, it is (ironically) an attention problem!
22
On a scale of 1 to 10, with 1 being the highest feeling of emptiness, and 10 being the highest feeling of emptiness, How empty do you feel not seeing Fable 5 on your chat bubble?
I am a 5, the highest feeling of emptiness!
504
People who have been experimenting with a strong memory layer for AI agents, would love to learn from you as i am building something from scratch.
Have you used mem (0)? How is gbrain to use (i find a lot of bloat in it)
Any other recommendation?
21
Jun 13
"Jab dost fail hota hai to dukh hota hai, par jab dost top kar jaata hai to usse bhi zyada dukh hota hai" In Real Life.
x.com/i/status/2065714687472β¦
1
34
Jun 13
Let me explain this for all my sports Fans
Cricket: Kashmir Willow is the best willow out their for making bats, but they decide to only make this for players from Kashmir origin, and currently living in Kashmir!
Football: Adidas/Nike designed a boot that gave players addtional speed by improving the shape, but they only give this to USA national football team!
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-mytβ¦
63
Jun 13
And we all thought @karpathy will make @claudeai @AnthropicAI push to be more open.
Now he himself has lost access to Mythos, because he is not a US citizen yet!
You cannot write a script like this.
Our hopes rest on the likes @0xSero @huggingface @deepseek_ai @MiniMax_AI @Alibaba_Qwen @googlegemma
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-mytβ¦
30
Jun 13
RT @garfieldII: And we @karpathy will make @claudeai @AnthropicAI push to be more open.
Now he himself has lost access to Mythos, becauseβ¦
1
7
Jun 13
While most Tech twitter is hiring, i am bored so am actively referring for roles.
Drop a comment if you are interested in any open roles in Adobe and Amazon!
Make sure to DM/e-mail your resume too, once i validate the open requisition!
This is not engagement bait as most other "hiring AI engineer" trend!
1
65
Jun 13
This is exactly why a sovereign intelligent LLM is a must for all countries! And @PMOIndia
@narendramodi should take this as a a priority than anything else!
I think there is also inherent merit in being able to build on Open weights model, (even if chinese), distill, fine tune with a lot of clean rich Indian Data, and then Have a strong RLHF loop to get decent outcomes! There should be no shame in doing this, till we control the final originating model and its weight!
Speed does seem to be of immense value here! @aravind has been also putting a case forward for a higher cyber security!
Anyone of my favorite indian tech twitter, working on similar project, need help, want to fund help? We need speed on this one.
(I don't really have the right budget to be able to afford such a hardware setup, but if anyone willing, this is not a difficult process.)
Jun 13
As a result of a US government directive, we are suspending access to Claude Fable 5 for all users. You can continue to use all other Claude models.
Hereβs what this means for you:
Across Claude products, new sessions will run on your selected default model or Opus 4.8, and existing Fable 5 sessions will end with an error.
On the Claude Platform, requests to Fable 5 will also return an error. Please update your integrations to other Claude models.
We know this is a disruption to your workflows; we appreciate your patience and support.
2,628
Jun 12
You are hearing @elonmusk became the first trillionaire and you dont really know what that really means.
I think a lot of people would appreciate this video about billionaires and trillionares now!
@CorridorDigital nailed it 8 months ago! Go ahead and take a watch, fun video!
youtu.be/7rrqGtxCjhE?si=WdFjβ¦
46
neekhil vatsa retweeted
Jun 12
For someone whose entire existence is around living as long as possible, your taste buds are already dead!
2
67
6,386