Pinned Tweet
After a major engineering effort at @elevenlabs, we're introducing Eleven Flash v2 & v2.5 - models that build on our previous Turbo series to deliver a new standard for ultra-low-latency performance.
We've brought inference time down from around 250ms to under 75ms, all while preserving human-like quality and keeping over 32 languages.
Maximiliano Levi retweeted
Big reasoning speedups are here. 1,000× faster training and 280× less compute than any looped model. All done using a fully trainable energy from the 80s
Keep reading to hear how energy is the right paradigm for reasoning
With @HughBlayney @jcrwhittington @prefrontal_ai
🧵/10
Maximiliano Levi retweeted
We raised $8M to build the software verification layer for the agentic coding era.
We hit 1.5M npm downloads, 12k GitHub stars and 60 contributors in the first month.
Backed by: @CredoVentures, @kaya_vc, @PointNineCap, @baobab_ventures and @scriptcapital.
Joined by incredible angels: Oscar Salazar (co-founder of Uber), @marcinzukowski (co-founder of Snowflake), @Carles_Reina, @chadbyers, @emilkowalski, @lukas_m_ziegler, @MattSwulinski, and many more.
Very proud of the whole team.
We're just getting started. 🫡
TesterArmy is now e2e.
The software verification layer for the agentic coding era.
Announcing our new brand identity and $8M seed round.
e2e.com/blog/seed
Maximiliano Levi retweeted
Today we are happy to share one of our first open research releases: Latent Sparse Attention (LSA), a step towards subquadratic attention that matches accuracy of full attention at ultra-long context.
Sparse attention attends to only a few tokens per query, but it has to find the right ones. At 10M tokens, that search takes 98% of the attention layer's time. LSA makes sparse attention up to 22.7× faster and full attention 72.4x faster.
With @kubicek_ales @caciolai @PengLiangzu @SukritSumant @Ben_Hoov @FraserGreenlee @RMBattleday @radim_urban @thoefler @jcrwhittington
60 year old plumber just came over, fixed my pump, flexed how rich he is and then told me AI was going to take my job
can't make this shit up
3
204
Maximiliano Levi retweeted
ElevenLabs' Eleven v4 Turbo takes the #1 spot on the Artificial Analysis Provider Voice TTS Arena Leaderboard, ahead of Eleven v4 while costing half the price
Eleven v4 Turbo is the faster, lower-cost sibling of Eleven v4 from @ElevenLabs, designed to deliver comparable speech quality with lower latency and cost.
Key takeaways:
➤ Provider Voice: Eleven v4 Turbo has an Elo of 1,334 (+19/-19) across 1,645 appearances, placing it ahead of Eleven v4 at 1,321, Alibaba's Qwen-Audio-3.1-TTS-Plus at 1,292 and Cartesia's Sonic 3.6 at 1,278. It ranks #1 in Customer Service and Entertainment and #2 in Assistants and Knowledge Sharing.
➤ Controlled Voice: Eleven v4 Turbo ranks #2 with an Elo of 1,173 (+15/-15) across 1,745 appearances, behind Qwen-Audio-3.1-TTS-Plus at 1,181 and ahead of Eleven v4 at 1,160. It ranks #1 in Arabic, German, Hindi and English (UK).
➤ Pronunciation Robustness: Eleven v4 Turbo scores 90.1%, ranking #2 behind Eleven v4 at 91.7% and ahead of Gemini 3.8 Flash TTS at 89.5%.
➤ Pricing: Eleven v4 Turbo is priced at $40/1M characters, half the price of Eleven v4 at $80/1M characters, compared to Sonic 3.6 at $49/1M characters and Gemini 3.8 Flash TTS at $16.49/1M characters.
➤ Speed: Eleven v4 Turbo processes 96 characters per second of generation time, compared to 84 characters per second for Eleven v4 and 59 for Eleven v3.
See more details and listen to samples below 🧵
Maximiliano Levi retweeted
A couple of things that make Eleven v4 special:
- Extremely low latency (100 ms p50 TTFB for Turbo)
- By far the best voice similarity
- Outstanding quality (#1 on Artificial Analysis and on our own benchmarks)
It achieves all of this while being extremely reliable. Eleven v4 uses our new post-training stack and a surprising recent research discovery that we hope will greatly accelerate our future model development.
By the end of the year, we expect delivery of new compute that will increase our training capacity almost 10x. Very excited for what's ahead!
Congratulations to the team!
Introducing Eleven v4 and Eleven v4 Turbo, our fastest and most emotive voice models yet.
Ranked #1 by Artificial Analysis.
Maximiliano Levi retweeted
This week, we brought all of @ElevenLabs together. 800 of us, from 50 countries!
Remote-first has been a superpower for us, but there’s nothing like having everyone in one place and feeling the energy across the team. Now onwards - so much more to build!
Maximiliano Levi retweeted
Time to scale that SSI:
We are announcing a long-term strategic partnership with NVIDIA. NVIDIA is making a substantial investment in SSI that will let us 10x our compute in the next 12 months. We reached the point where our research is worth scaling and with this partnership we will be able to. We are honored by NVIDIA’s conviction.
Maximiliano Levi retweeted
Replying to @tkexpress11 @midjourney
By VC standards we should either "conquer the world or die in a fire" and neither of these are spiritually compelling for me. I never wanted a company I just wanted a home. At this point we have a large and loyal paid community, we build tons of features for them (I think we did >30 releases this year) and they're pretty happy. We have enough revenue to fund tons of crazy R&D and our models are still the best by the metrics we care about (how the images look and how fun it is to make things). We have a huge backlog of exciting things to make our models way better. Zero risk. We did all this with no investors. Honestly, it feels like we are successful. The next metric of success I think about most about is now that we have "all I ever wanted" in terms of a big well funded R&D lab with cool people free to work on whatever we want... Can we now build something that would make baby David proud? And can we tell bold stories about a human future that people want to be a part of? I think we can.
at some point anthropic will start competing on verticals with it's own app (e.g. legal, vibe apps, health) and it's classifier will start nerfing those api requests too...
4
5
248
love this!
A new Claude launch felt like the right moment to share a work about #Claude’s first announcement.
“Introducing 1484 19223” is a 30 × 30 in. backlit lenticular light box from Lexicon 4178 44, a physical artwork series I created about tokenization in modern LLMs.
Viewed from different angles, the artwork shifts between the English words humans read and the token IDs Claude 1 processes: making tokenization visible as a physical process.
The colors are inspired by Claude Code’s color scheme, and the dark palette with LED backlighting evokes the glow of screens, terminals, and development environments where we now collaborate with AI.
Full series: anthonyluocreative.com/lexic…
@AnthropicAI @claudeai @ClaudeDevs
1
247
Maximiliano Levi retweeted
I created Lexicon 4178 44, a physical artwork series that reveals the hidden tokenization process within modern LLMs.
The first work, “Introducing 16047 38 2898,” is a 7-flip, 30 × 30 in. lenticular artwork based on the 2022 announcement of ChatGPT by @OpenAI. Viewed from different angles, the artwork shifts between the English words and GPT-3.5 token IDs: making tokenization visible as a physical process rather than an invisible preprocessing step.
The piece is showing in the #CVPR2026 Art Gallery in Denver, CO (curated by @elluba) starting today.
@cvpr #CVPRArt
Maximiliano Levi retweeted
We raised $500M at an $11B valuation to transform how people interact with technology.
great article!
Today Thinking Machines Lab is launching our research blog, Connectionism. Our first blog post is “Defeating Nondeterminism in LLM Inference”
We believe that science is better when shared. Connectionism will cover topics as varied as our research is: from kernel numerics to prompt engineering. Here we share what we are working on and connect with the research community frequently and openly.
The name Connectionism is a throwback to an earlier era of AI; it was the name of the subfield in the 1980s that studied neural networks and their similarity to biological brains.
thinkingmachines.ai/blog/def…
7
657
Maximiliano Levi retweeted
Introducing Eleven Music. The highest quality AI music model.
- Complete control over genre, style, and structure
- Multi-lingual, including English, Spanish, German, Japanese and more
- Edit the sound and lyrics of individual sections or the whole song
Maximiliano Levi retweeted
I left quant to build again.
Over the last few years, I’ve watched my friends build impactful companies like Cursor, Cognition, and Mercor from scratch. The signal is clear:
Model capabilities are accelerating exponentially—and the biggest opportunities now lie in applying them with precision.
Today, I’m thrilled to announce @averahq.
Avera is an applied AI research lab built on four core beliefs:
1. Step changes in model capabilities will continue
2. Every downstream industry will be transformed
3. Applied research at the frontier will unlock transformational software
4. Avera’s unique purpose is to bridge the best technology humankind has ever seen to industry
We’re a lean team of exited founders, national champions, PhDs, and elite engineers from Harvard, Columbia, Cambridge, and more.
It’s the best time ever to be a builder. We’re based in NYC and hiring exceptional talent across engineering, design, and research. Join us.
averalabs.ai























