AI Coding Live News

On November 9 at NVIDIA AI Day Seoul, LG AI Research made the K-EXAONE 2.5 API free through its EXAONE API Platform (docs.exaone.ai). Developers, startups, enterprises and research institutions can get an API key and use it for service development and research. K-EXAONE 2.5 is a technical validation model built by further training K-EXAONE 2.0, released in August: 750B total parameters with 31B active, text-only, Korean and English, a 512K context and an OpenAI-compatible endpoint. On four agentic benchmarks run internally, LG reports 59.2 on SWE-Bench Pro, 78.0 on BrowseComp, 74.6 on AutomationBench and 77.3 on Claw-Eval, roughly on par with DeepSeek-V4.1-Flash and GLM-5.3 but not leading across the board. NVIDIA says the model was trained with Nemotron datasets. LG has not published how long the free access lasts or any quota, and has not released 2.5 weights.

Zoom

Tencent published tencent/Qwen3.8-27B-DSpine-b16 on Hugging Face at 14:20 UTC on 2026-11-07: a five-layer speculative-decoding draft model of about 2B parameters with block size 16 and an 8,192-token sliding window, under Apache 2.0, with code at GitHub Tencent/DSpine and the paper at arXiv 2609.36173. The model card (vendor figures) shows that with SGLang 0.5.22, BF16 and concurrency 8, output throughput on a SWE-bench Pro replay at xhigh reasoning rose from about 248 to about 791 tokens/s (about 3.19x), versus 2.87x for the official DFlash2; acceptance length beat DFlash2 in all 10 agent settings. Note that DSpine drafts 16 tokens per block while DFlash2 drafts 8, so the setups are not identical, and no third-party reproduction exists yet.

Zoom

On 2026-11-05 Microsoft Foundry blogged that Foundry Local CLI 0.13.0+ can run 4-bit Qwen3.8-27B (~18 GB weights) on-device via ONNX Runtime and connect it to GitHub Copilot through Bring Your Own Key. Needs an NVIDIA GPU with at least 24 GB VRAM (about 64K context on 24 GB; up to the model's 262K max on 32 GB+). Vendor-cited model-card scores include Terminal-Bench 2.1 73.0 and SWE-bench Pro 61.7; Microsoft also reports higher generation speed than same-quant llama.cpp on RTX 5090 and DGX Spark. Mac and AMD support are still on the roadmap.

Zoom

Ricursive Intelligence released its first model, Ripple-65B Preview, on November 4 (18:24 UTC), open-sourcing the instruction-tuned model (BF16 and NVFP4) and a base model on Hugging Face under Apache 2.0. The official model card describes a 65B dense Transformer pretrained on 12T tokens with up to 1M context, and self-reports TerminalBench 2.1 80.5, SWE-bench Pro v2.0 Hard 79.1 and DeepSWE 1.1 44.3, ahead of much larger US open-weight models on agentic coding. It fits on local hardware such as the DGX Spark. CEO Anna Goldie says it cost about a tenth of the prior comparable state of the art.

Zoom

Shanghai AI Laboratory has moved Atria Dawn from Preview to a full release, with open weights (plus an FP8 build) on Hugging Face under internlm/Atria-Dawn and on ModelScope, MIT licensed. Built on the 744B-parameter MoE GLM-5.2, it targets long-horizon research and engineering agent work; context doubles from the Preview's 256K to 512K and input is text only. The vendor table shows it leading the listed models on AutomationBench, CyberGym and SkillsBench, while still trailing Claude Opus 5 on coding tests such as SWE-bench Pro and Terminal-Bench 2.1 (all vendor-reported).

Zoom

Recursive, founded by Richard Socher, has officially open-sourced Flint (RecursiveSI/flint), a 39.8B Mixture-of-Experts (MoE) coding agent model under Apache 2.0 with day-zero Ollama support (ollama run flint). Activating just 2.5B parameters per token across 47 MoE layers with a 262K context window, Flint hits 125+ tokens/sec on an M4 Pro MacBook Pro. It sets new benchmarks for sub-50B local models, scoring 74.2% on SWE-bench Verified and 89.7% on SWE-bench Pro v2.

Zoom

On Oct 21, Recursive, the startup founded by Richard Socher, released its first model, Flint: a 39.8B-total MoE coding model with about 2.5B active parameters per token, with weights on Hugging Face under Apache 2.0. Recursive says it leads sub-50B MoE models on SWE-Bench Pro v2, Terminal-Bench 2.1 and SWE-Bench Verified, runs at 125+ tok/s on an M4 Pro MacBook Pro, and was built from scratch in under four months with its automated research system, the Eureka Machine, contributing to nearly every part of the stack.

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

Meta AI published the technical report for Muse Glimmer 30B (arXiv:2610.24628), an open-source 30-billion parameter causal language model with an integrated perception encoder designed for on-device autonomous agents. Quantized to approximately 4 bits, Muse Glimmer operates inside 32GB of VRAM on a single consumer GPU (such as an RTX 5090) without cloud reliance. On rigorous agentic evaluations, Muse Glimmer sets new records for its size class: achieving 75.5 on MCPAtlas (surpassing Gemma4-31B at 54.2 and Qwen3.6-27B at 62.5) and 51.2 on SWE-Bench Pro, released fully under the Apache 2.0 license.

Zoom

Mainstream coding agent benchmarks like SWE-bench frame evaluation around issue-derived tasks: feeding an agent a bug description and expecting an end-to-end patch. But does this reflect genuine developer practices? JetBrains Research analyzes 4,782 production sessions of professional software engineers inside JetBrains IDEs (arXiv:2610.09633). For long interactive sessions (>=3 turns, comprising 33% of production traffic), user requests span a diverse mix of intent types—project codebase Q&A, strategic planning, code review, refactoring, and test execution—with users frequently transitioning between types mid-session. To align benchmarks with real developer task flows, JetBrains introduces SWE-TaskFlow, an approach that transforms issue-based benchmarks via prompt-splitting and verifiable repository QA into calibrated multi-stage workflows guided by a TaskFlow Alignment Score (TFAS). A pilot study across 700 SWE-Bench Pro tasks demonstrates that sequential multi-step interaction protocols nearly double agent inference expenditure without stable improvements in resolve rates, proving that the interaction protocol itself is a critical, previously unmeasured dimension of agent evaluation.

Zoom