Co-founder at @theforage_. Working with a great team to help our community of +2 million students build career skills through company-authored curriculums.
San Francisco, CA
Joined August 2020
- Tweets132
- Following712
- Followers247
- Likes1.4K
Tom Brunskill retweeted
We’re sharing more of our research, starting with our paper on outcome-routed post-training for diffusion language models.
Many standard RL methods are difficult to apply reliably to diffusion language models. Our approach avoids policy likelihood ratios and combines reward-based learning with teacher guidance.
Tool-calling accuracy (BFCL) rose from 53.55% to 88.6%, without increasing the model’s size.
The result: a more effective way to post train diffusion language models for new capabilities.
Our team will present the work at the NeurIPS 2026 DiffuLM. arxiv.org/abs/2610.19271
Tom Brunskill retweeted
Apparently AI has a 10% chance of wiping out humanity. But those enterprise SaaS targets aren't going to hit themselves.
Beyond the hype, the economics are changing rapidly. Open models are closing the gap, costs are falling, and intelligence is becoming increasingly commoditised.
The endgame for AI isn't just intelligence. It's the rate at which you can produce useful intelligence for a given amount of energy.
Had a great discussion with Pip Bingemann and @TomHums on the Wild Hearts podcast. We covered everything from AI doomerism and the concentration of intelligence, to why speed matters more than you think, how open source is changing the economics of AI, and why there's never been a better time to build!
Go listen (or watch)!
What becomes possible when AI gets fast enough?
For Celeris, speed is a research problem, not just an engineering optimization. When a model responds in milliseconds, entirely new classes of real time applications become possible.
Our co-founder and CTO @jn2clark on Wild Hearts, the @blackbirdvc podcast.
Tom Brunskill retweeted
Celeris-1 Decision was recognised as the fastest decision model by Artificial Analysis, by a factor of 2x.
It was placed on the speed/intelligence Pareto frontier, performing 4 points higher on the intelligence index than GPT-6-Luna.
@ArtificialAnlys benchmarks Celeris-1 Decision as the fastest decision model at 62ms median response time, 2x faster than @typesafeai Jev.
According to the benchmark, Celeris-1 Decision scores 4 points higher than @OpenAI's GPT 6 Luna.
Tom Brunskill retweeted
Excited to launch Celeris-1-decision, a multimodal diffusion decision model, which scores accuracy of 80% on Jev-Bench, 7% higher than Jev 1.13.0. It has improved latency relative to Jev, with a median response time of 67ms.
We tested Celeris-1-decision against 4 models, including Jev, Luna, Perplexity and Mercury Decide, and Celeris-1-decision came out ahead of all of them on Jev Bench.
Finally - it supports explanations, which means you not only get a decision back but also an explanation of why it chose that option.
Available at celeris.ai, $0.04 per million input tokens, $0 for output tokens.
Celeris-1 Decision is live today.
On jev-bench and typed decisions it beats Jev 1.13.0, OpenAI gpt-6-luna, Perplexity pplx-decider-v1.1-27b, and Inception Mercury Decide, on both accuracy and Brier score. Celeris-1-decision also led the benchmark in response time with a median of 67ms.
$0.04 per million input tokens, output free.
Full results and methodology: celeris.ai/celeris-1-decisio…
Tom Brunskill retweeted
Celeris-1 Decision is live today.
On jev-bench and typed decisions it beats Jev 1.13.0, OpenAI gpt-6-luna, Perplexity pplx-decider-v1.1-27b, and Inception Mercury Decide, on both accuracy and Brier score. Celeris-1-decision also led the benchmark in response time with a median of 67ms.
$0.04 per million input tokens, output free.
Full results and methodology: celeris.ai/celeris-1-decisio…
Tom Brunskill retweeted
Introducing Celeris-1 Magnus. A model built for agentic work.
On τ³-bench banking, Magnus delivers 41.2% at a 55-second median, against GPT-5.6-sol's 38.1% at 79 seconds.
It’s a hybrid diffusion model derived from qwen3.8-27b, optimized for agentic workloads.
For more info, check out: celeris.ai/celeris-1-magnus
Tom Brunskill retweeted
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.
Tom Brunskill retweeted
From the printing press to telecommunications, each major leap in human progress has been driven by an increase in information throughput. AI opens the next frontier: intelligence throughput—how quickly models can understand, decide and act.
Celeris-1 is an early proof of that direction. Built through diffusion, architecture search and AI-driven experimentation, it shows how new inference architectures can deliver frontier intelligence on a radically faster clock.
Proud of what this team has accomplished—and excited by where this research leads next!
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 celeris.ai
Tom Brunskill retweeted
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 celeris.ai
Tom Brunskill retweeted
Excited to launch Celeris-1 - a general purpose language model which delivers near-GPT-5 level intelligence with 15x faster response times.
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.
The model is available starting today. Sign up at 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 celeris.ai
Tom Brunskill retweeted
congrats to our portfolio company @theforage_ just announced a huge partnership with @pearson!!
Forage's job simulations will be embedded into Pearson's materials (including eTextbooks and curriculum) - combining power of Forage's 3 million students and Pearson's +18 million US students and 3,700 universities.
Tom Brunskill retweeted
One of my favorite interviews from last year was discussing the evolution of edtech with Tom
I haven’t come across an outsider to @theforage_ who has picked up the essence of what we’re building so quickly. The challenges, nuances, tensions and opportunity - all of it. Loved working on this with @sarthakgh
Tom Brunskill retweeted
Grand Junction is amazing for a number of reasons. Now we have one more to add to the list thanks to @theforage_.
Coming amid Women's History Month, the #CO03 city was recently named the 3rd best city for women to launch careers.
Congrats, GJ!
bit.ly/40v1swS
Tom Brunskill retweeted
IT’S HERE #JFFHorizons podcast episode #Entrepreneurial Ambition–Building to Scale with @MansfieldTB, @BrunskillTom
of @Forage, @Anthillhelps, @FutureFitAI, @jenamcgregor at @Forbes, @mallory_dwinal @craftedusystem
🎧: jfflink.org/3xGlUP1
Tom Brunskill retweeted
Here are 240+ FREE virtual internships you can take to get experience in tech. You'll learn and add projects to your portfolio - from design, development, engineering, product & project management, data analytics, tech sales, strategy, and consulting.
Tom Brunskill retweeted
Congratulations to @theforage_ for being awarded 'Startup of the Year' at the 2002 @UNSW Founders Awards last month 🏆
Co-founder @Pashpops started out as a student and today he's a coach and mentor to emerging founders 💞
More on @StartupDailyANZ: buff.ly/3OXlr2V
Tom Brunskill retweeted
Congrats to the legends at @theforage_ for winning @UNSW start up of the year! Looking good @Pashpops! @BrunskillTom @Telstraventures startupdaily.net/topic/event…?
Tom Brunskill retweeted
When co-creating content with employers to help candidates understand what a job is like, the challenge is to balance access with customization to expand reach + usefulness
@BrunskillTom tells me they have created a "career awareness" metric for that sarharibhakti.substack.com/p…
Tom Brunskill retweeted
You *need* to know founder Tom Brunskill if you’re interested in #career mobility - he’s an expert
@BrunskillTom
Great interview by Sar. A few gems:
“The recruitment process is broken. Rather than ‘hire then train,’ employers should use a ‘train then hire’ model.” – at Sao Paulo, Brazil
[New Post] My most comprehensive chat so far
I have been following this Australian company for while now. Enjoyed investigating this multi-layered business that sits at the intersection of multiple trends & product approaches.
I talked to @BrunskillTom sarharibhakti.substack.com/p…