Advancing open intelligence through efficient, accessible infrastructure for AI inference and post-training.
Joined May 2024
- Tweets512
- Following76
- Followers688K
- Likes465
Gradient retweeted
Can a coding agent choose the right model at every step?
Excited to share that our collaboration with @Gradient_HQ, TwinRouterBench, is accepted by NeurIPS 2026! 🎉 We route model calls at each agent step based on what the task needs next.
On 100 held-out SWE-bench Verified tasks, our router cut API cost by 53.1% vs. always using Claude-Opus 4.6, while resolving 75 vs. 74 tasks.
Paper: arxiv.org/abs/2605.18859
Code: github.com/CommonstackAI/Twi…
Gradient retweeted
The Bitter Lesson of Asynchronous RLHF:
Why Staleness Matters and How to Control It linkedin.com/pulse/bitter-le…
Gradient retweeted
This is a glimpse of where local AI is heading and we are glad to be part of it.
Really impressive work by all the teams involved @Gradient_HQ, @tryParallax, and @GA_agent_ai
A self-evolving agent + a 428B model + 3 Macs = ?
Your own AI lab.
We ran @MiniMax_AI M3 locally with @tryParallax, right on our desk.
Then @GA_agent_ai took over to create a 5-stock portfolio and write it to disk.
No cloud. No API bills. Nothing left the machine.
Wild to see a ~3K-line agent drive all this with a 400B+ model on local hardware.
Thanks to the GenericAgent and MiniMax teams for making local AI feel real.
Gradient retweeted
nvidia going all in on local ai.
here's our take: it shouldn't depend on which chip you bought.
sparks, macs, the 5090 already on your desk, we cluster across all of it and split your favorite model pipeline-parallel so it runs fully private and local.
Gradient retweeted
We're hiring at Gradient.
Building open-source environment infrastructure for our distributed RL training stack — reproducible, scalable to thousand-GPU runs
Looking for 1–2 RL Environments engineers / tech leads: You've designed verifiers, built sandboxes for agentic RL rollouts, or shipped RL training data pipelines that survived contact with real training.
Domain depth in math, code, agent, tool, or GUI is a plus. PhD not required.
Also hiring research interns: PhD / Masters students with hands-on RLHF / RLVR / GRPO / DPO / agentic RL experience. Open-source footprint matters more than paper count. Most intern roles convert post-grad. No age cap. Founding-team-level equity for the right people.
DMs open.
Gradient retweeted
Thrilled to see @tryParallax live in production on @Theta_Network.
This is exactly why @Gradient_HQ built Parallax: turning the world’s GPU mesh into a sovereign, distributed token factory.
Congrats on the milestone! 🫡
To make this work, we adapted Parallax, @Gradient_HQ's distributed inference framework, to run across EdgeCloud's global node network. One API endpoint, model split across many machines, no centralized cluster required.
Gradient retweeted
glad we could help!
with the agentic adoption soaring, privacy and token cost are already the top concerns for both agent and human users.
that's what parallax's built for.
To make this work, we adapted Parallax, @Gradient_HQ's distributed inference framework, to run across EdgeCloud's global node network. One API endpoint, model split across many machines, no centralized cluster required.
Gradient retweeted
If software no longer needs you to operate it, what does an “application” even mean?
That’s what we’re digging into at The Agentic Shift with panels, demos, and speakers from Google, PixVerse, MiniMax + more.
SF | Apr 8
Sign up here:
lu.ma/y07o6vuo
Gradient retweeted
local ai has picked up fast since openclaw dropped.
with the latest wave of small capable models, more people are running serious workloads on their own hardware.
if you missed this good local ai tutorial from @yacinelearning or want a refresher on how distributed scheduling actually works under the hood, it's worth the rewatch over the weekend!
I am continuing my adventure into distributed AI system with the parallax scheduling strat from @Gradient_HQ
in this 37min tutorial I go through:
- heuristic used to make scheduling tractable
- dynamic programming formulation
- filling GPU with water
- shoving them into shelves
Gradient retweeted
2026 is the year agents go beyond coding.
Autonomous agents + permissionless capital
→ the agent economy goes brrrr.
Solana Agent Economy Hackathon: Agent Talent Show
Build the skill that represents your agent.
Show the app that empowers their agents.
Prize Pool: $30,000
Co-hosts: @solana & @trendsdotfun
Sponsors: @BitgetWallet & @solana
(Please refer to the quoted article for the detailed requirements for Solana & Bitget Wallet Prize Pool)
Public Submission Process
1. Publish an X Article
Introduce your submission (include all the relevant links) and share any story behind it that you want people to know in an X Article.
2. Quote RT this Hackathon Announcement post, tag @trendsdotfun @solana_devs @BitgetWallet with a hashtag #AgentTalentShow, and include the link to your X Article from Step 1 in the same quote RT
All X Article submissions will be displayed in a dedicated section on @attentionvc. Build in public.
Time
March 11, 2026 2PM UTC until March 27, 2026 2PM UTC
Gradient retweeted
The wait is over.
Open your Clawbox now!
The easiest way to own your🦞agent.
Zero code. Just download, click, and go.
Grab yours: commonstack.ai/clawdbot?ref=…
Gradient retweeted
Proud moment for the team 🚀
Our colleague delivered an oral presentation at AAAI, presenting our work on multi-agent collaborative reward design.
This work shows how collective intelligence—coordinated specialist agents—can produce more robust, interpretable reasoning than monolithic models.
At Gradient, we believe intelligence is built together:
models × agents × community.
And this is not a one-off — we continue to present strong work at top-tier venues, pushing the frontier of multi-agent learning and alignment.
More to come.
#AAAI #rewardmodel #LLM
Gradient retweeted
we sped up distributed inference by up to 5x with decentralized speculative decoding.
many don't realize that AI models normally generate text one single word at a time, waiting for the network after every word.
speculative decoding changes this by using a "guess & confirm" system, similar to autocomplete.
how it's done:
1. draft locally (the guess)
instead of waiting for the network, a tiny, fast model on your device guesses the next few words instantly, without waiting for the network.
2. confirm remotely (the check)
the massive remote model doesn't generate from scratch; it just checks the draft. it looks at the guesses in a batch and says "yes, yes, no." you get multiple words in the time it usually takes to get one.
3. adaptive logic
dsd is smart. if the topic is creative, it lets the draft flow loose. if the topic is math or code, it checks more strictly. it balances speed and precision automatically so your inference almost feel instant.
find out more:
paper: arxiv.org/abs/2511.11733
blog: gradient.network/research/tu…
Gradient retweeted
we made distributed inference verifiable with <1% overhead.
verification is critical for any distributed system. in a trustless network, actors may swap your 70B model for a cheaper 8B one to cut costs.
until now, maintaining inference integrity meant either doubling your cost (redundancy) or exploding your latency (zkp).
we created veri: an on-chain verification layer light enough for high-throughput frameworks like Parallax. it hits the economic sweet spot through architectural elegance:
1. commit-sample-verify
we don't prove every step; we check a random slice using game theory. workers commit to their work before the audit. cheating becomes statistically irrational, allowing a 1% sample to secure the entire sequence.
2. simultaneous execution
inference and verification happen simultaneously on the same worker pool. we don't need a separate "verifier set", so compute utilization stays high.
find out more about the architecture and benchmarks:
paper: arxiv.org/abs/2509.24257
blog: gradient.network/research/ve…
Gradient retweeted
the era of living inference is here.
thanks to rl, models aren't just predicting patterns. they are reasoning. thinking time is the new scaling law. it demands compute density that distributed systems can handle elegantly.
for parallax, serving tokens is too basic. we are switching on distributed thinking.
the future isn't built on pretraining. it is about intelligence that lives in distributed ways.
Gradient retweeted
Huge congrats to the winners of the Parallax competition!
We were blown away by the creativity and technical depth of the submissions. From massive local clusters to distributed AI apps, you pushed Parallax in ways we didn’t expect.
Here are the 8 winning projects. 🧵
