Introducing PRAXIST Beta, your autonomous research team🚀 Define the objective, constraints, and what success looks like. PRAXIST discovers the path. From there, PRAXIST takes on the experimental research loop. Multiple Research Peers explore competing approaches in parallel, share useful findings, and build on accumulated evidence across experiments and generations.... In partner-provided environments, PRAXIST reached a 100% safe-landing rate in a rocket simulation and reduced accumulated error in an industrial SLAM system from 9.37 centimeters to 5.01 centimeters. From robotics control to quantitative finance, PRAXIST has delivered measurable advances across fundamentally different problems by combining a shared core research architecture with domain-specific tools, knowledge, constraints, and evaluators. More approaches explored. Faster iteration. Greater R&D capacity without proportionally increasing specialist headcount.show more

Starting today, PRAXIST is open source. Explore the code. Adapt it to your domain. Discover what comes next. Learn More: GitHub: Tech Report: Announcement:

PRAXIST assembles the team each problem needs: analyzers, explorers, innovators, exploiters, falsifiers, and combinators. Research Peers pursue competing directions in parallel. Shared memory connects their evidence. A PI Panel decides what to test next.

General-purpose coding agents are optimized to complete software tasks. Many research agents use a winner-takes-all loop: keep the best result and discard the rest. PRAXIST is purpose-built for cumulative experimental research. Every useful finding can shape what happens next.

Across the full 75-task MLE-bench, PRAXIST (with an open-source model) achieved: → 65.3% Gold Rate — 49 gold medals → 80% Any Medal Rate — 60 total medals → 44% more gold medals than Claude Code + Claude Opus 4.8 → $3,054 in token cost, compared with $38,370 for Claude Code + Claude Opus 4.8 — roughly one-twelfth the cost Against the stronger Claude Code baseline, PRAXIST delivered better benchmark results at 92% lower token cost. *The token cost in USD for the full 75-task benchmark suite, measured July 1–30, 2026.

From robotics control to quantitative finance, PRAXIST has delivered measurable advances across fundamentally different problems by combining a shared core research architecture with domain-specific tools, knowledge, constraints, and evaluators. Across more than 100 research and optimization tasks, PRAXIST has been applied to problems spanning over 50 application areas.

Honestly, what you're building surprised me in a good way :)

Seems like my agents will be busy tonight

The shared memory part is really interesting ngl.

good stuff, curious how the pi panel decides which direction deserves another experiment 🤔

The PI Panel looks beyond the current best result. If a direction shows a meaningful early signal but the evidence is still weak or noisy, it may run another experiment to validate, falsify, or better characterize the finding before deciding whether to pursue it further.

can i also wrap my project in this harness to make it work?

open source ftw 🙌

$3,054 versus $38,370 is a pretty serious cost gap for the same 75-task benchmark Curious how well it holds up once the research campaign gets much longer, i.e. for long-horizon tasks

This is a smart way to scale experimentation.

when a research peer’s direction fails, what survives into shared memory the raw result, or a distilled reason it failed? that’s where evidence compounds or turns into noise.

Shared memory carries findings rather than raw experimental outputs. PRAXIST captures what was learned, including relevant context around why a direction worked or failed, so other peers can build on the useful evidence without inheriting all of the experimental noise.

100% landings in sim is cool but how messy was the first gen before it got there

Open-sourcing PRAXIST makes this even more interesting.

Scaling R&D without scaling headcount is huge.

Autonomous research at this level is seriously impressive 🚀

PRAXIST is a really impressive research system

Curious where this starts to hit diminishing returns - does compute become the bottleneck first, or can the PI Panel keep getting useful search depth as the number of Research Peers scales?

Congrats on the launch!!

What was the hardest part of getting from 9.37 cm to 5.01 cm in the SLAM system?

This is huge team,congratulations and keep building!

Generality is the claim here, and a supplied sandbox cannot settle it. Whoever writes the evaluator writes the ceiling, so the honest test is an external benchmark nobody in the partnership designed.

Really interesting approach to autonomous research. The shared learning between research peers stands out.

The ability to run parallel research loops and build on previous findings is seriously impressive.

Congrats !!

This is a huge step toward scalable autonomous R&D.

thats some serious parallel research automation

Article:

Curious about the shared memory part. How much does carrying what the agents learned from earlier experiments actually improve the next research round?

Wow that's a good one. I thought it was a model but in fact it's a harness. Good job for the launch

better results. 92% cheaper. and they just open-sourced the whole thing.

Really curious how the results change if you keep the PRAXIST architecture but swap in a much stronger foundation model. Does performance keep scaling?

$3k vs $38k is crazy