Ravi Theja
@ravithejads
Followers
6K
Following
8K
Media
146
Statuses
1K
Applied AI @MistralAI Previously - @llama_index (LlamaIndex) Focused on LLMs, Agents, RAG, and fine-tuning LLMs.
Bangalore, India
Joined December 2010
๐ฅ Releasing BRAG: High-Performance RAG Model Trained In $25 ๐ Weโre thrilled to announce the launch of our RAG models, collectively known as BRAGโa series of SLMs (3 SLMs and 1 Ultra SLM) specifically trained for Retrieval Augmented Generation (RAG). 1๏ธโฃ โ BRAG-Qwen2-7b-v0.1
17
82
446
Mistral Vibe CLI running Devstral-small-2-2512 on M3 Ultra 512GB with LM Studio as backend ๐ฅ First part of the video 1x and the remaining 2x Great speed overall, 27 toks/s as average tracked on LM Studio console. Details to run it below ๐งต
10
36
301
New release from @MistralAI : Devstral 2 & Vibe CLI! 1๏ธโฃ Devstral 2 (123B): 72.2% on SWE-bench Verified, 256K context window, and 7x more cost-efficient than than Claude Sonnet, all under a modified MIT license. 2๏ธโฃ Devstral Small 2 (24B): 68.0% on SWE-bench, competing with
mistral.ai
State-of-the-art, open-source agentic coding models and CLI agent.
1
2
27
Introducing the Mistral 3 family of models: Frontier intelligence at all sizes. Apache 2.0. Details in ๐งต
175
833
5K
Super excited to share that I have joined the Applied AI, US team at @MistralAI. Grateful to @aviTwit3 and @sophiamyang for the opportunity, and thankful to the HR team, Charles, Alexandre, and Brian for making the transition seamless. A heartfelt thank you to @NirantK, Naveen
61
7
627
here is maya1, our open source voice model: Weโre building the future of voice intelligence @mayaresearch_ai team is incredible; amazing work by the team. remarkable moment.
149
341
3K
๐จHow do we improve long-horizon reasoning capabilities by scaling RL with only existing data? Introducing our new paper: "h1: Bootstrapping LLMs to Reason over Longer Horizons via Reinforcement Learning"๐ซก > RL on existing datasets saturates very quickly > Reasoning over
10
49
285
What-a-day, to be head-on competing against Elevenlabs, Apple and Google.. When I woke up today, I imagined to be the top launch of the day but post launch we have been laging behind the magnificient 7 by a bit. I need your support to compete against the biggies of tech, link
3
11
22
Introducing 20+ connectors powered by MCP and a fully controllable Memory in Le Chatโmaking it one of the most connected and relevant AI assistants for enterprises and consumers. Why switch to Le Chat? A ๐งต
55
218
1K
๐ฅ ๐๐ฎ๐ ๐ญ ๐ฟ๐ฒ๐น๐ฒ๐ฎ๐๐ฒ: ๐ ๐ถ๐๐๐ฟ๐ฎ๐น ๐บ๐ฒ๐ฑ๐ถ๐๐บ ๐ฏ.๐ญ Mistral Medium 3.1 is one of the top models in @rungalileo Agent Leaderboard and has an excellent cost<>perf tradeoff. Look at the avg turns and duration! Excellent release from @MistralAI.
2
5
36
is it possible to pretrain a language model using pure reinforcement learning from scratch? random weights, no cross-entropy loss pretraining. you may have many questions in your head.
63
160
2K
Donโt stop at context, engineer the workflow to solve agentic use cases. Prompt Engineering โ Context Engineering โ Workflow Engineering
Context engineering techniques are hot, but something I feel like is less talked about w.r.t AI agents is workflow engineering ๐ ๏ธ Deep research agents are extremely powerful with tool use but general - if you feed it a bad prompt, you wonโt get back good results. Your ability to
1
2
17
. @MistralAI Coding Stack for Enterprises is here! 1๏ธโฃ Codestral 25.08 โ +30% accepted completions & 50% fewer runaway generations 2๏ธโฃ Codestral Embed โ lightning-fast, high-recall search across huge codebases 3๏ธโฃ Devstral โ agentic workflows for multi-step development 4๏ธโฃ Mistral
mistral.ai
0
0
7
NEW @MistralAI Cookbook: @HubSpot Dynamic Multi-Agent System with @MistralAI Magistral Reasoning model: Ask questions in plain English โ Get analysis/ Report + automated CRM updates
2
8
40
Incredible feat by @sachdh Aryabhatta-1.0 scores 90.2% on JEE Mains, outperforming o4 mini and Gemini Flash 2.5. If you or your organization is looking for RL-as-a-service solutions, Sachin is the go-to person.
Excited to share Aryabhatta 1.0, our leading model that scores 90.2% on JEE Mains, outperforming frontier models like o4 mini and Gemini Flash 2.5 Trained by us at @AthenaAgentRL , in collaboration with @physics__wallah, using custom RLVR training on 130K+ curated JEE problems
1
1
15
.@MistralAI releases Voxtral, open source speech understanding models built for real-world applications Available in two sizes ๐น Voxtral 24B for production-scale deployments ๐น Voxtral Mini 3B optimized for local and edge use Voxtral delivers state-of-the-art transcription,
3
2
27
Introducing Devstral Small and Medium 2507! This latest update offers improved performance and cost efficiency, perfectly suited for coding agents and software engineering tasks.
92
324
2K
Excited to see @MistralAI Small 3.2 powering the text LLM behind Unmute, the open-source voice AI from @kyutai_labs ๐ค๐ค
Kyutai TTS and Unmute are now open source! The text-to-speech is natural, customizable, and fast: it can serve 32 users with a 350ms latency on a single L40S. Try it out and get started on the project page: https://t.co/B4P9FuOrQc
1
13
70
๐ Bridging Offline & Online RL for LLMs ๐ ๐: https://t.co/G12TS6Z84n New paper shows on verifiable & non-verifiable tasks: - Online DPO & GRPO give similar performance. - Semi-online (iterative) DPO with sync every s steps (more efficient!) works very well also. - Offline DPO
1
98
455
Today, Maya Research is launching Veena A speech model that speaks like India does. Every voice tells a story. Open source.
304
531
4K
Introducing Mistral Small 3.2, a small update to Mistral Small 3.1 to improve: - Instruction following: Small 3.2 is better at following precise instructions - Repetition errors: Small 3.2 produces less infinite generations or repetitive answers - Function calling: Small
55
230
2K