PRAXIST Beta is an autonomous research and development system from Singapore-headquartered artificial general intelligence research company Sapient Intelligence , aimed at engineering, ML, robotics, manufacturing, finance, and drug discovery teams that spend most of their time iterating through experiments. Users specify an objective, constraints, and a budget, and PRAXIST then autonomously experiments and evaluates the strongest solution . Under the hood, PRAXIST deploys multiple autonomous research peers to explore approaches in parallel, test hypotheses, investigate failures, and validate promising results , rather than executing a predefined path chosen by the user.
The launch matters now because R&D has largely resisted the productivity gains AI has delivered elsewhere, and Sapient is releasing benchmark numbers to make a concrete case. On 75 challenging Kaggle competitions from MLE-Bench, a benchmark designed to test how well AI systems tackle complex, real-world machine learning problems, PRAXIST achieved the highest-level result in 49 competitions at an approximate recorded model cost of US$3,000, compared with 34 highest-level results for Claude Code at approximately US$38,000 under the same evaluation conditions. For teams working with sensitive IP, PRAXIST can operate with proprietary data in private or customer-controlled environments, giving organizations greater control over their research infrastructure and sensitive intellectual property , and Sapient has signaled plans to extend it from technical R&D into broader operational optimization like inventory, logistics, and sales.
Sapient Intelligence was founded by Guan Wang and William Chen in 2024 , with research centers in San Francisco and Beijing alongside the Singapore headquarters. Co-founder William Chen frames the product this way: "PRAXIST is designed for long-horizon R&D, where the problem-solving approach itself may need to be discovered and adapted." The beta is the first public preview of the system, with an open-source release signaled to follow.
Comments (13)
Sign in to join the discussion.
N
Niloofar RazaviAug 28, 2026
'Multiple Research Peers explore competing approaches in parallel' is a great line until my cloud bill sees it. What's the token discipline here?
P
Pranshu V.Aug 28, 2026
Curious how PRAXIST handles reward hacking across the Peers. Is there a critic loop or are they just YOLOing gradient signals at each other?
C
Chidi OkonkwoAug 28, 2026
ok wait, autonomous research team means I finally get to ignore Slack for a full afternoon? Sold on the vibes, need to see it flail on a real problem though.
M
Marisol VegaAug 28, 2026
The tweet cuts off mid sentence and that's honestly the most human thing about an AI launch this week.
T
Tomas LindqvistAug 28, 2026
Rewriting the tagline for free: 'Smaller models. Sharper thoughts.' You can Venmo me later.
D
Deepa IyerAug 28, 2026
How does this compare to the HRM paper y'all put out earlier? Same architectural DNA or a different beast under the hood?
K
Kwesi MensahAug 28, 2026
hot take: 'autonomous research team' is the new 'AI agent' is the new 'copilot'. Show me one experiment it ran that a grad student wouldn't have tried first.
G
Grigor AshotyanAug 28, 2026
Parallel Peers is a cute story until you look at retention past week four. What does the second month of usage look like on the private beta?
V
Valentina CazaresAug 28, 2026
Before I forward this internally: SSO, SOC2, and where does the experimental data actually live? Procurement will ask before I finish my coffee.
B
Benoît AristideAug 28, 2026
Portfolio flashback: this feels like when we backed a certain lean-model shop that everyone said was too small to matter. Watching this one closely.
R
Ritika KaulAug 28, 2026
Who's on the research side over there? I have a systems person leaving a frontier lab who'd eat this stack for breakfast.
S
sundayfogAug 28, 2026
A research team that never asks for a coffee break is either a miracle or a monitoring nightmare.
M
Maddox CheruiyotAug 28, 2026
Where's the API doc link. Rate limits, webhook on experiment completion, and can I stream Peer logs or do I have to poll like it's 2014?
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Sapient Intelligence
Building leaner architectures for deeper intelligence.
Building leaner architectures for deeper intelligence.
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
PRAXIST Beta is an autonomous research and development system from Singapore-headquartered artificial general intelligence research company Sapient Intelligence , aimed at engineering, ML, robotics, manufacturing, finance, and drug discovery teams that spend most of their time iterating through experiments. Users specify an objective, constraints, and a budget, and PRAXIST then autonomously experiments and evaluates the strongest solution . Under the hood, PRAXIST deploys multiple autonomous research peers to explore approaches in parallel, test hypotheses, investigate failures, and validate promising results , rather than executing a predefined path chosen by the user. The launch matters now because R&D has largely resisted the productivity gains AI has delivered elsewhere, and Sapient is releasing benchmark numbers to make a concrete case. On 75 challenging Kaggle competitions from MLE-Bench, a benchmark designed to test how well AI systems tackle complex, real-world machine learning problems, PRAXIST achieved the highest-level result in 49 competitions at an approximate recorded model cost of US$3,000, compared with 34 highest-level results for Claude Code at approximately US$38,000 under the same evaluation conditions. For teams working with sensitive IP, PRAXIST can operate with proprietary data in private or customer-controlled environments, giving organizations greater control over their research infrastructure and sensitive intellectual property , and Sapient has signaled plans to extend it from technical R&D into broader operational optimization like inventory, logistics, and sales. Sapient Intelligence was founded by Guan Wang and William Chen in 2024 , with research centers in San Francisco and Beijing alongside the Singapore headquarters. Co-founder William Chen frames the product this way: "PRAXIST is designed for long-horizon R&D, where the problem-solving approach itself may need to be discovered and adapted." The beta is the first public preview of the system, with an open-source release signaled to follow.
Comments (13)
'Multiple Research Peers explore competing approaches in parallel' is a great line until my cloud bill sees it. What's the token discipline here?
Curious how PRAXIST handles reward hacking across the Peers. Is there a critic loop or are they just YOLOing gradient signals at each other?
ok wait, autonomous research team means I finally get to ignore Slack for a full afternoon? Sold on the vibes, need to see it flail on a real problem though.
The tweet cuts off mid sentence and that's honestly the most human thing about an AI launch this week.
Rewriting the tagline for free: 'Smaller models. Sharper thoughts.' You can Venmo me later.
How does this compare to the HRM paper y'all put out earlier? Same architectural DNA or a different beast under the hood?
hot take: 'autonomous research team' is the new 'AI agent' is the new 'copilot'. Show me one experiment it ran that a grad student wouldn't have tried first.
Parallel Peers is a cute story until you look at retention past week four. What does the second month of usage look like on the private beta?
Before I forward this internally: SSO, SOC2, and where does the experimental data actually live? Procurement will ask before I finish my coffee.
Portfolio flashback: this feels like when we backed a certain lean-model shop that everyone said was too small to matter. Watching this one closely.
Who's on the research side over there? I have a systems person leaving a frontier lab who'd eat this stack for breakfast.
A research team that never asks for a coffee break is either a miracle or a monitoring nightmare.
Where's the API doc link. Rate limits, webhook on experiment completion, and can I stream Peer logs or do I have to poll like it's 2014?