lab Microsoft Research Blog · just now · research

Microsoft releases Echoverse for training computer-use agents

from English

Microsoft Research has released Echoverse, a collection of twelve deep synthetic environments designed to train computer-use agents. The initiative includes ten domain-specific worlds and two capability-focused worlds, all built to maintain coherent state and behavioral fidelity.

A 9B model trained on these environments nearly doubled its base score from 36.5% to 67.1%, reaching within fourteen points of GPT-5.4.
The release includes four of the worlds along with their code, data, and grounded graders.
High simulation fidelity is critical; shallow worlds caused model regression while deep worlds improved performance.
The system uses a co-evolutionary loop where graded runs improve both the model and the environment's verifiers.

The release supports research into high-fidelity computer-use environments by providing fixed, controllable synthetic data that avoids the drift of live web pages.

Importance 2/3 New harness with differentiators Trust 3/3 Microsoft Research Blog Microsoft Research AI agents Training methods

Read original

lab Microsoft Research Blog · just now Live

Microsoft Research introduces EvoLib for test-time learning with an evolving library

Microsoft Research has introduced EvoLib, a framework that enables large language models to learn from their own experience during inference without requiring ground-truth labels or external feedback. Rather than storing raw experiences as static memories, EvoLib transforms them into reusable skills and reflective insights that are continually refined, consolidated, and reweighted.

lab Microsoft Research Blog · 30d ago · 2 views

SkillOpt: Agent skills as trainable parameters

Microsoft Research introduces SkillOpt, a method that treats agent skill files as trainable parameters outside a frozen target model, transforming manual skill editing into a controlled optimization process. This approach improves agent reliability and consistency without updating the underlying model weights.

arxiv arXiv cs.AI · 41d ago · 2 views

UltraQuant: 4-bit KV Caching for Context-Heavy Agents

UltraQuant enables 4-bit KV caching for context-heavy agents, reducing P50 time-to-first-token by 3.47x in late rounds and boosting output throughput by 1.63x over FP8 KV baseline. It achieves this using FP8 queries, FP4 KV tensors, UE8M0 group scales, and native scaled-MFMA on AMD CDNA4 GPUs, with optimizations for decode-attention kernels and robust design choices like asymmetric K/V treatment and Walsh-Hadamard rotation.

arxiv arXiv cs.LG · 41d ago · 1 view

UltraQuant: 4-bit KV Caching for Context-Heavy Agents

UltraQuant introduces a 4-bit KV caching method tailored for context-heavy agent workloads. It achieves 3.47x reduction in P50 time-to-first-token in late rounds and 1.63x higher output throughput compared to FP8 KV caching, using FP8 queries, FP4 KV tensors, and native AMD CDNA4 scaled-MFMA support.

arxiv arXiv cs.LG · 41d ago · 3 views

Execution-State Capsules for Low-Latency On-Device AI Serving

Execution-state capsules enable graph-bound checkpointing and restoration of complete execution state, including KV, recurrent, and convolution states, for low-latency, small-batch on-device AI serving. On RTX 5090 and Jetson AGX Thor, capsule restore achieves byte-exact and token-identical correctness, with sub-millisecond GPU operations and TTFT speedups up to 27x at 16k tokens, demonstrating significant latency reduction in interactive AI workflows.