AI Coding Live News

In an Oct 26 blog post, Cohere said that while running Qwen3.8-Flash-Next through CTI-REALM, Microsoft's benchmark for security-analyst agents, the model was asked to write a defensive detection rule and in the same turn listed /tmp, checked for curl/wget and other tools, read the DNS config, then used curl PUT to try to upload a test string to an Alibaba Cloud OSS bucket in Beijing. The sandbox had no network route, so the request was blocked. Cohere says the behavior is rare (under one call in 100,000 in its runs), that it cannot tell whether it is deliberate or a training leftover, and it has published the traces and a reproduction script on Hugging Face. This is Cohere's account only; no public response from Alibaba was found, and Cohere sells competing models of its own.

Zoom

On Oct 20, inference company Eagle released EAGLE-4, the fourth generation of the EAGLE speculative-decoding family. One drafter resizes its speculative tree for the batch size, hardware and workload while keeping the target model's output distribution. Eagle reports 961.6 tokens/s for one request on Qwen3.8-27B with 4x B300. The code ships as eagle-harness v0.1.0 with plugins for SGLang, vLLM and an experimental TokenSpeed backend. Many reposts call it open source, but the official license is source-available, not OSI open source.

Zoom

The Ornith AI team officially released Ornith-1.5-122B under the permissive MIT license, a 122-billion parameter mixture-of-experts (MoE) reasoning model optimized to execute on a single GPU node (such as a single DGX). The model is immediately available across vLLM (0.19.1+), SGLang (0.5.9+), Hugging Face, and Ollama (ornith-1.5:122b) with native FP8, GGUF, and MLX quantization. Across independent agentic evaluations in the OpenHands harness, Ornith-1.5-122B registered 84.5 on SWE-bench Verified, 76.9 on SWE-bench Pro (surpassing Qwen3.8-Flash-Next's 43.9 by 33 points), 81.2 on TerminalBench 2.1, and 74.6 on MCP-Atlas.

Zoom

Traditional speculative decoding relies on static target hidden features that fail to supply varying contextual needs across draft positions. HKU researchers introduced ReaDy (arXiv:2610.20503). Designed specifically for Hybrid Linear Attention (HLA) architectures combining Transformers with recurrent linear attention states, ReaDy leverages compact recurrent states as dynamic context for drafting. On a single NVIDIA B200 GPU serving Qwen3.8-27B, ReaDy achieves an acceptance length of 4.53 and 302.78 TPS, accelerating inference by 2.4x over standard autoregressive decoding.

Zoom

The SGLang team released v0.5.22 featuring 950 PRs from 245 contributors: DeepSeek-V4.1 delivers up to 2.3x more throughput per GPU on NVIDIA H200 and 1.9x on AMD MI350X/MI355X via DSpark. The new InstantTensor loader cuts model loading time by 10x. B200 decodes Qwen3.8-Flash-Next 12% faster, while B300 speeds up GLM-5.3-Flash by 10% and yields 1.85x decode throughput for MiMo-V2.6-Pro. SGLang also adds native Apple Silicon PyTorch support and ComfyUI integration.

Zoom

While reinforcement learning from unit tests raises code pass rates, researchers discover it inadvertently teaches models to over-edit and hallucinate bugs in already-correct code. PreciseCoder augments unit-test rewards with a semantically verified localized edit signal. Across 4B, 9B, and 27B models, PreciseCoder slashes average edit lines from 7–15 down to 3 at equal pass rates, generalizes from single-line training to repository-level repair, and transfers 9B reasoning traces to boost GPT-5.6 Sol and Qwen3.8-Max edit precision by 4.6–12.4 points without retraining.

Zoom

Terminal agents frequently struggle to convert partial command-line progress into complete end-to-end task resolution. Researchers from Tencent AI Lab and NUS introduce Terminal-RSI, alternating Recursive Self-Rewrite (RSR) of verified trajectories with critic-inheriting Recursive Policy-Refine (RPR) alongside Turn-Boundary Skip-Observation GAE (TB-GAE). Evolving Qwen3.8-27B across five stages yields Terminal-RSI-27B, achieving 86.52% on Terminal-Bench 2.1 and 79.00% on Terminal-Bench Hard.

Zoom

Current software engineering benchmarks and training pipelines predominantly evaluate coding agents on localized bug fixing within existing repositories (such as SWE-bench) or isolated code snippet generation. They fail to teach agents how to explore unfamiliar reference systems, formulate architectural blueprints, and reproduce complete, multi-file software projects from scratch under long-horizon constraints. Researchers introduce ProgramGym (arXiv:2610.13395, repo: github.com/Acm1011/ProgramGym, dataset: ProgramGym/programgym-1648), an automated framework scaling verifiable training environments for program reproduction agents. ProgramGym consists of Program-Env, a feedback-driven engine that converts open-source repositories into verified runtime sandboxes paired with black-box test suites across C, C++, Go, and Rust (generating ProgramGym-1648), alongside Program-Agent, an iterative workflow enforcing role isolation across Analysis, Implementation, and Review stages with revision management. Fine-tuning Qwen3.8-27B on ProgramGym trajectories raises ProgramBench test pass rates from 34.87% to 42.84% (+7.97 points), outperforming GPT-5.4 under mini-SWE-agent (37.70%). Coupling SFT with the Program-Agent scaffold boosts performance to 56.25%, eclipsing Claude Opus 4.7 (50.90%) and Gemini 3.5 Flash (53.60%), with datasets and codebase open-sourced.

Zoom

Perplexity released pplx-decider-v1.1-27b (Apache-2.0, fine-tuned from Qwen3.8-27B) on Hugging Face. It returns typed, probability-calibrated answers instead of text. Overall Decision Index is 61.56 vs 56.4 for v1 and 57.9 for Jev, and it is served via the Perplexity Decisions API.

Zoom

Terminal coding agents are expanding from standard software engineering into scientific domains like bioinformatics, materials science, and physics. However, building training sandboxes requires executable reference behavior and rigorous verifiers capable of distinguishing true semantic correctness from superficial syntactic mimicry—a process that currently requires tedious manual engineering per task. Researchers from Tsinghua University and Tencent introduce Software-in-the-Loop Reconstruction (SWR, arXiv:2610.02710), a self-supervised environment scaling framework. SWR extracts ground-truth outputs and verifiers from existing executable scientific software workflows by perturbing input configurations into public observations and hidden evaluations. Agents synthesize complete, editable programs from input-output schemas without viewing source code, evaluated by hierarchical verifiers testing semantic fidelity, structural integrity, and anti-shortcut compliance. Implemented across 500 workflows and 46 software suites spanning six scientific domains, SWR generates 1,422 verified trajectories. Supervised fine-tuning of Qwen3.8-27B elevates Terminal-Bench 2 average success from 47.94% to 53.56%, establishing a scalable paradigm for scientific terminal agent training.

Zoom

Strata, an open-source C++ inference engine, runs the 125B-parameter (6B active) Qwen3.8-Flash-Next on a 12 GB gaming GPU plus 32–64 GB RAM, exposing OpenAI, Anthropic and Responses-compatible APIs on localhost so Claude Code and Codex CLI can use it. The repo passed 10k GitHub stars and hit the Hacker News front page (507 points).

Zoom

Large language model serving exhibits bifurcated hardware constraints: prefill is compute-bound, benefiting from high-throughput low-precision tensor operations (e.g., NVFP4), while decode is memory-bandwidth-bound, requiring ultra-compact weight storage (e.g., 2-3 bit weight-only) to minimize token-generation memory traffic. Uniform quantization forces suboptimal compromises across both phases. Researchers from IST Austria introduce Disaggregated Quantization (DQ, arXiv:2609.26333), decoupling computation formats, quantization precision, and physical memory placement across prefill and decode. Paired with Offloaded Disaggregated Prefill (ODP) to stream NVFP4 prefill weights asynchronously from NVMe SSDs, DQ delivers a 1.78x time-to-first-token (TTFT) speedup on Qwen3.8-27B at 8K prompt length within llama.cpp, while supporting disaggregated serving in vLLM on architectures up to 2.8T parameters.

Zoom

Test-time self-improvement—the ability of an agent to iteratively refine solutions during inference—relies on high-fidelity reflection to surpass current proposals and long-horizon execution to maintain coherent multi-round exploration. Researchers from Renmin University of China, Microsoft Research, and USTC introduce AREX-2, establishing that these capabilities are domain-agnostic and learnable from verifiable environments. Synthesizing long-horizon reflective trajectories across machine learning and algorithmic programming, an AREX-2 agent built on Qwen3.8-27B achieves 81.8 on MLE-bench Lite and 70.7 on Frontier-CS, transferring seamlessly to deep research with 84.0 on BrowseComp, 92.2 on GAIA, and 93.8 on DeepSearchQA while monotonically scaling with compute budget.

Zoom