Vish Srivastava

Patna lad Living in the Bay Area, USA! Ecommerce is my bread & butter 😊. Selling Ecommerce software for 20 Years. Citizen of 🇬🇧🇳🇿 🇮🇳 and GC 🇺🇸

[jesse @jn2clark]

Fantastic to see Celeris-1 reach #1 on the @ArtificialAnlys speed leaderboard!

2,064 tokens/sec — over 30× the median for comparable models.

We haven’t scratched the surface of what becomes possible when useful intelligence arrives this fast. It’s not just that existing tasks become faster. It opens up a new paradigm where every transaction, click, and interaction can be interpreted and acted upon in real time.

Thank you to everyone using @Celeris_ai and sharing feedback. The team is working around the clock as we continually improve it!

[Tom Hamer @tom_w_hamer]

Celeris-1 now ranks #1 of all AI models on speed, according to the latest Artificial Analysis benchmark 🚀

@ArtificialAnlys

[Celeris @Celeris_ai]

Celeris-1 is now ranked as the #1 fastest AI model according to @ArtificialAnlys.

2,038 output tokens per second, #1 of all 591 models in response time, TTFT and time per intelligence task. 2x faster tokens/second than the next fastest model. In addition, we beat Grok 4.3 on HLE.

Celeris-1 proves it’s possible to deliver useful intelligence 10x faster than today's systems, without custom silicon.

[Celeris @Celeris_ai]

Introducing celeris-1.

A general purpose language model which delivers near-GPT-5 level intelligence with 15x faster response times.

We’ve developed a new inference architecture that uses diffusion techniques instead of conventional autoregressive generation - unlocking dramatically better speed while maintaining frontier-level intelligence.

Celeris-1 delivers a p50 response latency of 157ms - around 15× faster than GPT-5-mini and 17× faster than GPT-5 - while scoring comparably 76% on MMLU-Pro, compared with 78% and 81%, respectively.

On tokens per second, Celeris-1 achieves a throughput of 1,280 tokens per second vs 144 for gemini-3.5 flash-light using a reconstructed version of @ArtificialAnlys's tokens per second benchmark dataset.

The model is available starting today. Sign up at https://t.co/99ywRD88KS

[Celeris @Celeris_ai]

Celeris-1 advances our mission of maximizing useful intelligence delivered per unit of time.

How we built it, full benchmarks, and what comes next: https://t.co/kw4XQPbGu0

[Steve Krouse @stevekrouse]

The new @Celeris_ai model is fast! I love fast

[jesse @jn2clark]

Latency changes what software can be. Building Marqo made that clear: instant search enables experiences that simply do not work when users have to wait.

Today, @tom_w_hamer and I are launching @Celeris_ai , an AI research lab building the world’s fastest LLMs. We’ve been working on this for the past few months and are really excited to finally share it!

AI progress has largely been measured by capability. In real-time systems, speed becomes part of that capability. Voice must keep pace with conversation, coding tools with the people using them, and agents with every step they take. At scale, small delays compound across every document reviewed or transaction processed.

At @Celeris_ai , our objective is to maximize intelligence delivered per second. We’re developing new model architectures and inference systems designed to deliver useful intelligence at extremely low latency.

Our first model is coming soon.

Read more at https://t.co/Y8ebCADVm8 .

[Vish Srivastava @vish_uk]

@tom_w_hamer @Celeris_ai @jn2clark Brilliant job Jesse and Tom 🎉

[Tom Hamer @tom_w_hamer]

Today we’re launching @Celeris_ai , an AI research lab building the world’s fastest LLM.

More than four years ago, @jn2clark and I set out to build Marqo. Today, it powers 10s of billions of dollars in global ecommerce revenue.

Jesse and I have decided to launch a second, even more ambitious project, alongside @marqo_ai . A team of engineers and researchers has been working at Celeris for the last few months, and today we're sharing it publicly for the first time.

Our focus at Celeris is on algorithmic and engineering advances that deliver frontier intelligence at microsecond response times.

Real-time voice, interactive coding, autonomous agents, robotics, and scientific infrastructure all require models that can respond in milliseconds or microseconds. Today builders of these systems must settle for smaller, weaker models, or ration frontier intelligence because it can’t keep up.

Our objective is simple: maximize useful intelligence delivered per unit of time.

Model launch coming soon - read more at https://t.co/WXZWiTHPcO

[Vish Srivastava @vish_uk]

#NewProfilePic

[Vish Srivastava @vish_uk]

California designates
Diwali as an official statewide holiday #thankyou

[Vish Srivastava @vish_uk]

My flaky forecast is that @elonmusk will be the first non US born President of the US. BTW, I am apolitical. 🇺🇸🥰

[Vish Srivastava @vish_uk]

#NewProfilePic

[Vish Srivastava @vish_uk]

@elonmusk @elonmusk I/we can help with the search. @Lucidworks search engine is the only one which has a scalability to index Twitter contents with right/relevant search results. Please let me know whom to speak in your org. Thank you

[Vish Srivastava @vish_uk]

@elonmusk @Lucidworks can help! Shall we pitch you or your team member the search solution we have? It will be the 2nd best decision you will make in Twitter takeover.

[Vish Srivastava @vish_uk]

Be healthy and happy and also look out for those who need your help! Be kind always 🙏🏽 Happy Diwali 🪔

[Vish Srivastava @vish_uk]

RIP! QueenElizabeth

https://t.co/x0z4Z8kUia

[Vish Srivastava @vish_uk]

#NewProfilePic

[Vish Srivastava @vish_uk]

Every @IPL team should have one @ICC associate country player in it. It will do great to there respective teams. @SGanguly99 @BCCI @JayShah @virendersehwag @imVkohli @nassercricket @MichaelVaughan

[Vish Srivastava @vish_uk]

Do you remember when you joined Twitter? I do! #MyTwitterAnniversary