Post-training Runs
We train frontier models on Surge datasets, then study what they learn: benchmark gains, transfer beyond the training distribution, and the behaviors that emerge.
We trained a model on 1,700 coding tasks.
It got better at office work.
Training on them improved GDPval-AA from 30.7% to 37.6%, even though GDPval-AA contains no coding tasks. Full blog post coming soon.
We trained a model on DAYJOB-style office work.
It got better at coding.
Training on Enterprise Agents: Frontier Work, long-horizon finance, tax, accounting, and healthcare assignments with no coding tasks, improved Terminal-Bench 2.1 from 67.4% to 75.8%. Full blog post coming soon.
GDPval success doubled after training on 1,400 GDP.pdf tasks
We post-trained Kimi K2.7 on 1,465 GDP.pdf companion tasks, where it worked on one document at a time with no tools. The GDP.pdf holdout full-task pass rate rose from 11.0% to 24.5%, and on GDPval, success more than doubled, even though GDPval requires navigating a workspace and using tools to dig into multiple files. The trained model used tools to find more of the relevant sources before creating deliverables, while reading less text overall.
We trained a model on 1,700 coding tasks. It learned skills that generalized across every SWE eval.
Training on 1,700 expert-built coding tasks improved Kimi K2.7 across every external SWE eval we measured. The biggest behavioral changes showed up in the complexity of real-world software engineering: carrying messy specifications through, testing beyond the obvious path, preserving existing behavior, and building its own checks when the environment did not provide one.
We trained a model on office work RL environments with no code. It got better at coding anyways.
We trained on long-horizon office tasks with no coding examples. SWE-Bench Pro still improved +5.7pp. Trajectory analysis pointed to a broader capability we call goal-directed execution: choosing useful subgoals, maintaining task state, staying faithful to the objective, and verifying completion.
We trained a model on single-turn ComplexConstraints. It also got better at multi-turn instruction following.
Every ComplexConstraints training example was single-turn, yet the largest gains appeared in multi-turn behaviors. Our hypothesis is that the underlying skill is shared: maintaining many interacting constraints without dropping them, whether they arrive in one dense prompt or accumulate over nine turns.
We trained a model on long-horizon tool-use environments. It learned agentic behaviors that transferred across every tool eval.
We trained Qwen3.5-122B-A10B on realistic, multi-tool agent environments. The gains transferred to Toolathlon, τ²-Bench, and BFCL-V4. Trajectory analysis showed better task closure and more efficient tool use across unfamiliar evaluation environments.
We trained a model on expert STEM problems. The gains transferred across every scientific reasoning benchmark.
We trained GLM-4.7 on expert-authored STEM problems. It nearly doubled performance on held-out Expert STEM tasks, with gains also appearing on HLE, FrontierScience, PubMedQA, and DeepDive. The result suggests that difficult expert problem solving can teach reasoning patterns that remain useful across different scientific domains.
We trained a model on customer-support RL environments. It learned tool-use skills that generalized beyond.
We trained GLM-4.6 inside CoreCraft, a high-fidelity customer-support environment. The gains transferred beyond CoreCraft to BFCL, τ²-Bench Retail, and Toolathlon. The result suggests that realistic enterprise workflows can teach general agent skills that survive changes in tools, tasks, and evaluation environments.
See what high-quality datasets teach your models
All of these experiments use Surge datasets available off the shelf. Run the training data on your model, measure what changes, and see whether the capabilities transfer.
Train on the full dataset first. Pay only if it moves the metrics that matter.