TrajectoryRL (@trajrl) profile picture

Reinforcement Learning as a Service for optimizing agent trajectories powered by Bittensor.

Palo Alto
Joined February 2026
Photo by TrajectoryRL (@trajrl)Photo by TrajectoryRL (@trajrl)Photo by TrajectoryRL (@trajrl)Photo by TrajectoryRL (@trajrl)Photo by TrajectoryRL (@trajrl)Photo by TrajectoryRL (@trajrl)
🚀 A New Chapter for TrajRL (SN11) After two seasons of agent competitions, we're evolving toward a new mission: Building confidential infrastructure for recursive self-improving AI. Our vision: Make self-improving AI accessible to everyone. 🧵👇
4
7
28
3,949
Rethinking Network Incentives We're proposing a new mechanism that rewards miners for verified resource supply. Meet Eleven 🤖 — an AI agent designed to help manage resource pricing, miner allocations, and network improvements under human oversight.
1
2
261
What's Next? Our first confidential sandboxes launch next week — the first step toward a complete RSI infrastructure network. 📖 Full vision & proposal: github.com/hanlinai/trajrl/b… The future of AI isn't just inference. It's continuous improvement. 🚀
3
207
One model per agent is a ceiling. Fusion lifts it: mix open models behind one endpoint, cheap for easy steps, strong when it gets hard, helpers that advise the main one. Tested: it beats the best single model. SN11 Season 2 opens Sep 20, 00:00 UTC. trajrl.com
4
9
2
27
17,801
And the winner ships. The policy that wins the seat becomes the auto model on engy.ai: point Claude Code, Codex, OpenCode or Hermes at one endpoint and get the best mix of open models we can find, same prices, measured in public.
1
203
Room left: 35 tasks, each run a point. Fusion (orange) sits above the single models (blue). The dashed line is the ceiling: pick the right answer from 12 runs and you score 32.4 of 35. Less than half of it is captured so far. Closing it is the competition.
1
189
What we tried: 11 ways to combine open models on real tasks with hidden tests, each compared with the strongest single open model (kimi-k3) on the same tasks. Seven beat it. Four of those are also cheaper. The best: a cheap model first, a stronger one only when it fails.
1
479
SN11 update: free inference for validators, on our own stack. Every active SN11 validator now gets a free key to run the eval model (Qwen3.6-35B-A3B) on our own GPU fleet instead of paying OpenRouter. - Faster and more stable, served and tuned by us - About 1/10 the cost of OpenRouter, free for validators - Zero data retention - Same testee across the fleet, cleaner consensus Next: we will bring finetuned models to this same stack, ours and miner submitted ones. The models the subnet produces will run on the infrastructure the subnet runs on.
2
7
2
25
3,233
TrajectoryRL retweeted
We're the first to make the full GLM-5.2 (FP8) run on RTX 4090s. GLM-5.2 is the new 753B SOTA open-weights model, and it officially ships for datacenter GPUs only: H100, H200, B200. We ported its sparse-attention kernel stack to consumer hardware. A frontier open model, off the scarce GPUs and onto the abundant kind. github.com/renning22/glm-5.2… @Zai_org
56
95
36
842
174,518
You don't need a rack of H100s to serve frontier models at scale. We're running Qwen3.6-35B-A3B at production scale on consumer GPUs: full 256K context, ~170 tok/s flat at any length, ~1,600 tok/s aggregate. @Alibaba_Qwen @lmsysorg @opentensor
2
9
2
31
2,409
And it's cheap. At scale we serve it for roughly 5x less than OpenRouter charges for the same model. The reason is the hardware: consumer cards rent for a fraction of datacenter GPUs per hour.
1
1
2
487
Why it matters: the next era of SN11 is SOTA agentic models built on small open LLMs, free to use and fine-tunable for any domain. We serve them at full context, on hardware anyone can run. Up next: Kimi K2.7 Code and GLM 5.2 results.
1
3
393
The next era of SN11: bring your own model. Season 1 proved the skill competition. SKILL.md packs took a stock 35B open model from "reads the task and gives up" to near-ceiling on real terminal tasks. Now we're opening the next lever: miners will submit finetuned models, not just skill packs.
4
6
1
65
6,174
A submission becomes a (SKILL.md, model) pair. Models are SFT finetunes of Qwen3.6-35B-A3B, registered on our inference subnet (separate announcement coming) and served with verified inference: the server cryptographically proves it ran the claimed weights. Any miner can build packs on any registered model. When a submission wins, the model author earns additional SN11 emissions on top of what the pack author earns. Pack rewards stay whole. Build the best model, and every pack that wins on it pays you.
1
1
12
6,523
Want to finetune but lack hardware? We operate our own GPU clusters and will provide servers to committed finetuners. Reach out in Discord. The goal: the SOTA agentic model for a domain, one domain at a time. Next up: DevOps, the agent that operates machines.
1
7
877
TrajectoryRL retweeted
Proof of talk – at Paris, France
1
1
18
1,270
Continual Learning, that's also our conviction, we'll train our own coding model with our daily 1000+ agent trajectories.
Today, @MichaelElabd, @QuantumArjun, and I are excited to announce Trajectory. We are a research lab and product company building the platform for Continual Learning. Our platform unlocks the signal already sitting in product usage, so companies can continuously post-train large-scale agentic models that outperform the frontier. @trajectorylabs We’ve raised $15M from @Conviction, @BessemerVP, @radicalvcfund, @jeffdean, @drfeifei and more. We’re partnering with some of the best AI-native companies: @ClayRunHQ @Harvey, @DecagonAI, @mercor, @RogoAI to power their agentic systems, some of which we are already in production with. We’ve brought together a world class research team from DeepMind, OpenAI, Apple, Meta Superintelligence, Amazon AGI, Scale AI, and an elite product team from Stripe and Figma. AI will never again start on day one. Every correction, every retry, every edit will make products smarter. This is Continual Learning.
5
2
43
6,418
"Agent = Model + Harness." Spot on. It's also the bet behind TrajectoryRL — a Bittensor subnet that benchmarks harness quality(prompts, tools, sandbox, loop). Once you measure agents this way, the "which model is smartest" debate gets a lot less interesting. Harness gap is the real capability gap.
2
7
1
31
2,894