AI Coding Live News

At 12:11 UTC on 2026-10-25, the Nanbeige LLM Lab of BOSS Zhipin (Kanzhun Limited) announced Nanbeige4.3-33B from its account @nanbeige (post), with the model and its Base version released as open weights on Hugging Face under Apache 2.0 (model card). It is a 33B dense model on a Llama-style architecture with a 512K context window, positioned as the teacher model of the Nanbeige4.3 series that will be distilled into smaller models such as Nanbeige4.3-3B. The official model card compares it with Qwen3.8-27B, Muse-Glimmer-30B and Gemma4-31B using the same harness at highest thinking effort: 80.9 on Terminal-Bench 2.1 (Qwen3.8-27B 73.0) and 65.5 on DeepSWE (42.5), but 60.3 on SWE-Bench Pro, slightly below Qwen's 61.7, and 84.8 on GDPval vs 86.4. All figures are self-reported and not yet independently reproduced.

Long-horizon coding agents tackling complex software repositories depend on timely guidance, yet conventional LLM critics often degrade agent outcomes by misinterpreting in-progress exploratory commands or offering ephemeral advice that is never tracked. Researchers from Rutgers and Lehigh University unveil Opera (arXiv:2609.33987), an open-source verbal critic framework that treats diagnostic corrections as persistent notes tracked until root problems are definitively resolved. Opera orchestrates reviews through hybrid periodic and event-driven triggers, verifies proposed critiques against visible terminal artifacts, and monitors downstream trajectories to distinguish mere superficial compliance from genuine resolution. As a test-time intervention, Opera elevates baseline agent resolve rates by 12.4 percentage points on Terminal-Bench 2.1, 15.0 on SWE-Bench Pro, and 8.9 on DeepSWE v1.1 across four foundation backbones. Furthermore, fine-tuning Qwen3.5-9B on Opera-guided rollouts yields a 10.2 percentage point gain on held-out repositories without any test-time critic, preserving stability across heterogeneous execution harnesses (such as OpenHands to Terminus-2).

Cline said Tencent Hy4 Preview currently leads SWE-bench Pro. The open-weight model is described as 770B parameters with 49B active and a 1M-token context window, and Cline called it Tencent's biggest generational leap measured to date. The practical developer path is immediate: npm i -g cline, run /model, and select Hy4 preview. SWE-bench Pro is a harder, more realistic software-engineering eval than classic HumanEval-style completion, so a lead there is a stronger signal for repository-level agents than a short coding quiz. The 1M context window is the other half of the story: long-horizon refactors, multi-file patches, and eval traces can stay in a single session instead of being retrieved in fragments. Cline's distribution also matters; putting Hy4 beside GLM-5.3 and Qwen3.8 in the same agent makes model shopping a /model command rather than a new IDE.

Alibaba Qwen released open-weight Qwen3.8-Flash, a multimodal Mixture-of-Experts model billed as an early preview of the Qwen4 architecture. The production-scale checkpoint is 125B parameters plus 51B N-gram embeddings, with only 6B activated per token. The new stack combines Gated DeltaNet plus Qwen Sparse Attention, Gated Residual branches, N-gram embeddings that can live in host memory, and a refined Muon optimizer. Qwen said it trained at about one-ninth the cost of Qwen3.7-Plus while beating it across the board, especially on coding and office work: 58.7 on DeepSWE 1.1, 62.5 on SWE-bench Pro, 73.9 on CoWorkBench, 84.5 on AndroidWorld, and 95.7 on MathVision with CI. Native context is 262K tokens, extendable to 1M with YaRN; at 1M context the QSA kernel is up to 7.6x faster in prefill and 4.9x faster in decode. QwenCloud lists roughly $0.15 input / $0.47 output per million tokens, and day-0 runtimes arrived from NVIDIA NeMo, SGLang, Unsloth, OpenRouter, and OpenCode Go.

The SWE-bench benchmark team launched SWE-bench Pro, expanding from Python to Rust, Go, C++, and TypeScript. Featuring 1,200 curated real-world issues from high-traffic open-source repositories, it provides a unified industry standard for multi-language autonomous agent software engineering evaluation.

Augment Code rebuilt the Auggie CLI harness on top of a minimal Pi fork, reducing schema overhead to achieve equivalent SWE-bench Pro performance at $1.27 per task (a 53% cost reduction versus Claude Code).

SambaNova launched hosted inference for MiniMax M3 on SambaCloud, highlighting its 1M-token context window, 59.0% score on SWE-Bench Pro, and optimized RDU hardware acceleration.

Alibaba Qwen open-sourced Qwen3.8-Flash, a multimodal MoE with 125B parameters plus 51B n-gram embeddings and only 6B active per token. It scores 58.7 on DeepSWE 1.1 and 62.5 on SWE-bench Pro, was trained at 1/9 the cost of Qwen3.7-Plus, and ships Qwen3.8-Flash-Next as an early look at the Qwen4 architecture.

Alibaba Qwen open-sourced Qwen3.8-Flash, a 125B-parameter multimodal MoE activating only 6B parameters per token. Delivering 62.5 on SWE-bench Pro and 58.7 on DeepSWE 1.1 at 1/9 the training cost of Qwen3.7-Plus, it previews the Qwen4 architecture with 262K native context (1M with YaRN) and low API pricing ($0.16 / $0.47 per 1M tokens).