Tech Meridian ← ENTITY INDEX
PROMY MERIDIAN RU

COMPANY · ENTITY #11771

Qwen3.8-27B

Related event timeline, sources and context from the news index.

EVENT TIMELINE

9

COMPANIES · 1 SOURCE · The Decoder

Nous Research's Hermes Agent adds 'Collective Wisdom' and Microsoft Teams support

Nous Research says its open-source Hermes Agent can now run inside Microsoft Teams (desktop app also in the Microsoft Store) using Azure Container Apps Sandboxes. A new 'Collective Wisdom' feature turns successful agent workflows into reusable team skills that, according to Nous, remain inside the company and are not used to train external AI providers; Nous also says Hermes is the most-used open agent on OpenRouter, processing nearly two trillion tokens per day.

7.0

MODELS · 1 SOURCE · Hugging Face

Hugging Face trains RL LoRA harness adapters for Qwen3.8-27B to boost performance and efficiency

Hugging Face trained LoRA adapters for Qwen3.8-27B inside three agent harnesses (pi, OpenCode, Hermes Agent) using RL with TRL's AsyncGRPOTrainer, OpenEnv and Harbor. Each adapter improved Terminal-Bench 2.1 performance and reduced token usage under its target harness; the team also merged the three LoRAs into a single super-adapter and published all artifacts and recipes on the Hugging Face Hub.

7.0

MODELS · 1 SOURCE · Hugging Face

RadixArk releases DaisyRose-27B, a Qwen3.8-27B-based RL-post-trained 27B model

RadixArk released DaisyRose-27B, a 27B-parameter model built on Qwen3.8-27B and post-trained with reinforcement learning using their Miles framework on long-horizon agent tasks (sessions up to 1M tokens). Trained over 18 days on 144 H200 GPUs using ~1,200 curated tasks, the model raises Terminal-Bench 4.1 pass@1 from 5.6% to 12.0%; checkpoints are available under Apache 2.0 and the team recommends serving with SGLang and DSpark speculative decoding.

7.0

RESEARCH · 1 SOURCE · The Decoder

Co-Design Bench finds coding agents fully complete only ~26% of real user tasks

Researchers (led by Alibaba and several Asian universities) released Co-Design Bench, a new benchmark using real user change requests for web apps; across ten models the best system (GPT-6 Astra) met about 70% of requirements on average but produced a fully correct result in only 26.4% of tasks, meaning human cleanup is still needed on nearly three out of four tasks. The study also shows performance depends strongly on task complexity and the agent harness (execution environment), so authors recommend reporting model and environment together.

8.0

RESEARCH · 1 SOURCE · The Decoder

Adobe Research and UC San Diego introduce EditAgent, an AI that chooses specialized editing tools and verifies edits

EditAgent, from Adobe Research and UC San Diego, decomposes image-editing requests into steps, pits a general-purpose editor against specialized tools for each step, and uses a standard language model fed verification records (close-ups, change maps, and measured statistics) to pick the best variant. On a 600-edit benchmark it outperformed prior agent judges and improved several open editors, but needs roughly ten times the compute of a single editor; the verified outputs also produced useful training data.

7.0

RESEARCH · 1 SOURCE · Hugging Face

ThinkingBox-Training released on GitHub; RL fine-tune raises Qwen3.8-27B reliability using database-graded rewards

ThinkingBox-Training is now available on GitHub. The toolkit turns executable checks of terminal database state into RL rewards; fine-tuning Qwen3.8-27B with 50 GRPO steps raised average pass@1 from 51.70% to 60.78% and increased tasks solved on all 20 attempts from 38 to 107, using an offline policy-relative difficulty selection of training tasks and evaluation on the 507-task ThinkingBox-Bench.

7.0

COMPANIES · 1 SOURCE · NVIDIA Developer

NVIDIA adds official eight-node DGX Spark clusters with 1 TB unified memory to run frontier models

At GTC Berlin NVIDIA announced official support for eight-node DGX Spark clusters, scaling unified memory to 1 TB and enabling local execution of frontier models including the full GLM 5.3 (753B) from Z.ai. NVIDIA also introduced a Spark-optimized collective backend called RoCEnante (merged into FlashInfer) that reduces all-reduce latency—reporting a 4.43x decode speedup at eight nodes (vs 3.05x without it)—and published benchmarks (e.g., Qwen3.8-27B NVFP4 scaling from ~12 to 52 tokens/sec, GLM 5.3 ~80 TPS, and 7.79x pretraining speedup at 97.4% efficiency for a nanochat model); NeMo Switchyard and vLLM integration are highlighted for multi-model routing and distributed inference, and NVIDIA Sync eight-node support is due later this year.

7.0

MODELS · 1 SOURCE · The Decoder

Cloudflare launches Clef and Clef-flash decision models (Qwen-based)

Cloudflare released two decision models, Clef (based on Qwen3.8-27B) and the smaller Clef-flash (Qwen3.5-9B), which output probability-weighted choices for predefined answers so downstream agents can act automatically. The company says both support text and images, report median latencies of ~209 ms (Clef) and ~39 ms (Clef-flash), are compatible with TypeSafe AI's Jev API, available on Hugging Face under Apache-2.0, and will be accompanied by a reinforcement-learning service for customer fine-tuning.

8.0

RESEARCH · 1 SOURCE · arXiv cs.AI

AREX-2 paper introduces long-horizon reflective training for self-improving LLM agents

AREX-2 (arXiv:2609.38288v1) describes training LLM agents built on Qwen3.8-27B using synthesized long-horizon improvement trajectories from machine-learning and algorithmic programming tasks to teach reflection and sustained iterative improvement. The paper reports strong results (MLE-bench Lite 81.8; Frontier-CS 70.7) and transfer to research benchmarks (BrowseComp 84.0; HLE 52.6; GAIA 92.2; DeepSearchQA 93.8), with performance that continues to improve as iteration budget grows.

7.0