WorkPublished All news from that day
Prime Intellect says over 2,000 agents rewrote Prime Agent in Rust in two weeks
Prime Intellect's own account says over 2,000 agents did the rewrite in two weeks, with parity tests as the check and humans fixing what the tests missed.
Check our sources · 12 facts from 1 sourcePrime Intellect published its account of rewriting Prime Agent, its open-source coding agent, in Rust on 9 October 2026. It says Prime Agent, launched in August, has been downloaded more than 300,000 times and has processed over 8 trillion tokens.
The company says that over two weeks Prime Agent orchestrated a swarm of over 2,000 agents to rewrite itself, working across more than 10,000 Prime Sandboxes and using over 200 billion tokens from the GLM-5.3 endpoint of Prime Inference.
A single root agent split the work into tasks, and each task passed through a planner, an implementer, a reviewer on a different model and a verifier that ran parity checks in a fresh sandbox.
The checks compared the TypeScript and Rust versions on terminal output, session transcripts, protocol messages and a feature audit. The company says the checks only verified the behaviour they exercised, and that its internal team found bugs and missing behaviour while using the Rust build every day. It says humans were in the loop for finding problems, directing changes and reviewing results.
A second loop ran for three days to improve speed and memory use, and the company says it merged over 69 changes. It reports time to input roughly 14 times faster than the TypeScript version and over 80% less memory after startup, measured on its own benchmark suite, and cautions that comparisons with other agent harnesses should be read with care.
Your reaction
We count reactions per story and day, never who reacted. The counts help us choose what goes in the monthly issue. If you are signed in, your own page shows yours too.
We count reactions per story and day, never who reacted. They help choose what goes in the monthly issue.
Check our sources
Every sentence above is checked against this source.
1 Rewriting Prime Agent in Rust
Open the source archived copy-
Prime Intellect's post "Rewriting Prime Agent in Rust" carries the publication date 2026-10-09.
citeRewriting Prime Agent in Rust ... OCT 9TH, 2026
-
Prime Intellect says that since it launched Prime Agent in August, the agent has been "downloaded more than 300,000 times and has processed over 8 trillion tokens".
citeSince we launched Prime Agent in August, it has been downloaded more than 300,000 times and has processed over 8 trillion tokens.
-
Prime Intellect says: "Over two weeks, Prime Agent orchestrated a swarm of over 2,000 agents to rewrite itself end to end, operating across 10,000+ Prime Sandboxes with over 200 billion tokens from Prime Inference’s GLM-5.3 endpoint."
citeOver two weeks, Prime Agent orchestrated a swarm of over 2,000 agents to rewrite itself end to end, operating across 10,000+ Prime Sandboxes with over 200 billion tokens from Prime Inference’s GLM-5.3 endpoint.
-
Prime Intellect says a single root agent orchestrated the rewrite and divided the work into a topological ordering of tasks, and that the root agent wrote no product code.
citeA single root agent orchestrated the rewrite and divided the work into a topological ordering of tasks. The root agent wrote no product code
-
Prime Intellect says each task moved through four agents: a planner, an implementer, a reviewer that inspects the pull request "using a different model and in a separate context from the implementer", and a verifier that runs the feature's parity checks and tests in a fresh Prime Sandbox.
citeeach task assigned by the orchestrator moved through four agents: Planner ... Implementer ... Reviewer: inspects the pull request adversarially, using a different model and in a separate context from the implementer ... Verifier: compiles and runs the feature's parity checks and tests in a fresh Prime Sandbox.
-
Prime Intellect lists four kinds of parity check: TUI parity, harness parity, protocol parity and feature parity.
citeTUI parity: a differential test suite compares the TypeScript and Rust binaries side by side
-
Prime Intellect says: "the parity checks only verified the behavior they exercised", and that moving its internal team onto the Rust build "exposed bugs and missing behavior outside the coverage of the differential tests".
citethe parity checks only verified the behavior they exercised ... which exposed bugs and missing behavior outside the coverage of the differential tests.
-
Prime Intellect says that after the rewrite, "Prime Agent still wrote and tested these changes, while we had humans in the loop involved in finding problems, directing the changes, and reviewing all results."
citePrime Agent still wrote and tested these changes, while we had humans in the loop involved in finding problems, directing the changes, and reviewing all results.
-
Prime Intellect says a second orchestrator ran a hillclimbing loop for three days, logged over 144 experiment and audit records and merged over 69 valid changes that boosted performance.
citea second orchestrator ran a hillclimbing loop for three days ... the loop logged over 144 experiment and audit records, merging over 69 valid changes that boosted performance.
-
Prime Intellect says, on its own custom runtime suite, that time to input is roughly 14x faster than its TypeScript version and memory after startup over 80% lower; the "amongst the fastest" ranking is its own and runs only attributed to it.
Toned down to what the source says
citeWith time to input roughly 14x faster than TypeScript and using over 80% less memory after startup, Prime Agent is amongst the fastest coding agent harnesses available.
-
Prime Intellect says its external harness results were measured with a custom runtime suite and that "Without a common benchmark standard, comparisons should be interpreted with caution."
citeExternal harness results were measured with our custom runtime suite. Without a common benchmark standard, comparisons should be interpreted with caution.
-
Prime Intellect says the rewrite ships with native Windows support (beta) and installation through homebrew, and that Prime Agent remains open source.
citewe are also releasing Prime Agent with native Windows support (beta) and installation through homebrew. Prime Agent remains open source
Every sentence above is checked against this source. Each fact has its own address you can cite.
Source 1 · Prime Intellect · 9 Oct 2026
Rewriting Prime Agent in Rust
Prime Intellect's post "Rewriting Prime Agent in Rust" carries the publication date 2026-10-09.
Rewriting Prime Agent in Rust ... OCT 9TH, 2026
Prime Intellect says that since it launched Prime Agent in August, the agent has been "downloaded more than 300,000 times and has processed over 8 trillion tokens".
Since we launched Prime Agent in August, it has been downloaded more than 300,000 times and has processed over 8 trillion tokens.
Prime Intellect says: "Over two weeks, Prime Agent orchestrated a swarm of over 2,000 agents to rewrite itself end to end, operating across 10,000+ Prime Sandboxes with over 200 billion tokens from Prime Inference’s GLM-5.3 endpoint."
Over two weeks, Prime Agent orchestrated a swarm of over 2,000 agents to rewrite itself end to end, operating across 10,000+ Prime Sandboxes with over 200 billion tokens from Prime Inference’s GLM-5.3 endpoint.
Show all 12 factsShow fewer facts
Prime Intellect says a single root agent orchestrated the rewrite and divided the work into a topological ordering of tasks, and that the root agent wrote no product code.
A single root agent orchestrated the rewrite and divided the work into a topological ordering of tasks. The root agent wrote no product code
Prime Intellect says each task moved through four agents: a planner, an implementer, a reviewer that inspects the pull request "using a different model and in a separate context from the implementer", and a verifier that runs the feature's parity checks and tests in a fresh Prime Sandbox.
each task assigned by the orchestrator moved through four agents: Planner ... Implementer ... Reviewer: inspects the pull request adversarially, using a different model and in a separate context from the implementer ... Verifier: compiles and runs the feature's parity checks and tests in a fresh Prime Sandbox.
Prime Intellect lists four kinds of parity check: TUI parity, harness parity, protocol parity and feature parity.
TUI parity: a differential test suite compares the TypeScript and Rust binaries side by side
Prime Intellect says: "the parity checks only verified the behavior they exercised", and that moving its internal team onto the Rust build "exposed bugs and missing behavior outside the coverage of the differential tests".
the parity checks only verified the behavior they exercised ... which exposed bugs and missing behavior outside the coverage of the differential tests.
Prime Intellect says that after the rewrite, "Prime Agent still wrote and tested these changes, while we had humans in the loop involved in finding problems, directing the changes, and reviewing all results."
Prime Agent still wrote and tested these changes, while we had humans in the loop involved in finding problems, directing the changes, and reviewing all results.
Prime Intellect says a second orchestrator ran a hillclimbing loop for three days, logged over 144 experiment and audit records and merged over 69 valid changes that boosted performance.
a second orchestrator ran a hillclimbing loop for three days ... the loop logged over 144 experiment and audit records, merging over 69 valid changes that boosted performance.
Prime Intellect says, on its own custom runtime suite, that time to input is roughly 14x faster than its TypeScript version and memory after startup over 80% lower; the "amongst the fastest" ranking is its own and runs only attributed to it.
Toned down to what the source saysWith time to input roughly 14x faster than TypeScript and using over 80% less memory after startup, Prime Agent is amongst the fastest coding agent harnesses available.
Prime Intellect says its external harness results were measured with a custom runtime suite and that "Without a common benchmark standard, comparisons should be interpreted with caution."
External harness results were measured with our custom runtime suite. Without a common benchmark standard, comparisons should be interpreted with caution.
Prime Intellect says the rewrite ships with native Windows support (beta) and installation through homebrew, and that Prime Agent remains open source.
we are also releasing Prime Agent with native Windows support (beta) and installation through homebrew. Prime Agent remains open source
Topics
The morning email
On the mornings we publish: the three top stories and up to four short ones. Free.
We email you a link to confirm. An issue may include one sponsor, always labelled Sponsored · Advertisement. Our emails count opens and clicks, not who made them. Unsubscribe in one click. What we keep