Latest news, beside the leaderboard

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.

Perplexity officially unveiled that Perplexity Computer now operates 100% locally on macOS through the native desktop application. Breaking away from cloud-dependent computer-use paradigms that ship continuous video feeds across the internet, Perplexity Computer executes both the underlying inference model and the agent orchestration runtime directly on local Mac silicon. By interfacing natively with macOS Accessibility APIs, the system eliminates network latency, guarantees complete screen privacy, and empowers autonomous desktop copilots to orchestrate multi-application workflows entirely offline.

Researchers introduced HERA in arXiv:2610.06563, addressing a critical reliability gap in autonomous LLM agents: the inability to detect task infeasibility and reliably abstain from blind, destructive tool execution. Rather than tuning harnesses against static benchmarks that quickly saturate, HERA establishes a harness-environment co-evolution framework that mutates execution environments to expose emergent failure modes while adapting the scaffold. Evaluated across 19 frontier LLMs using the newly released hera-bench, HERA elevates agentic abstention success from 61.7% to 83.3% and paired task success from 48.3% to 70.0%.

While supervised fine-tuning (SFT) distills agentic tool-use capabilities from frontier LLMs into small language models (SLMs, 0.6B to 8B parameters), student agents frequently suffer from 'progress stagnation'—collapsing into repetitive, unproductive tool-calling loops as interaction trajectories extend. In arXiv:2610.19439, researchers introduced Progress-Aware Distillation (ProD). ProD defines a rigorous stagnation scoring formulation that dynamically identifies and penalizes stagnant behavior during student trajectory rollouts via an adaptive SimPO-inspired margin. Experiments across 8 diverse SLMs prove ProD effectively eradicates repetitive loops while boosting task completion efficiency and success rates.

DeepSeek officially open-sourced DeepInfer under Apache 2.0, its production-grade inference engine serving DeepSeek-V4.1-Flash at scale. Originating from vLLM with architectural refinements from SGLang and TensorRT-LLM, DeepInfer natively targets NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs. It integrates specialized kernels including FlashMLA, DeepGEMM, DeepEP, and TileKernels. The engine features DSpark confidence-scheduled speculative decoding with 5-token drafts, SWA bounded replay over a 140-token sliding window, and balanced MoE expert parallelism. DeepSeek simultaneously open-sourced deepseek-recipe, delivering an end-to-end open ecosystem across models, kernels, and inference runtimes.

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.

Researchers introduced MemT-1.0 (arXiv:2610.26825), a groundbreaking architecture internalizing memory as an intrinsic component of foundation model computation. Moving beyond exploding KV caches and lossy Vector RAG pipelines, MemT-1.0 employs Gated DeltaNet-2 with stable within-turn token-level updates alongside a novel Mixture-of-Storage mechanism. Trained across a two-stage 34B-token pipeline, the released MemT-1.0-Nano model achieves a 1.86x end-to-end inference speedup and a staggering 66x reduction in persistent storage at 128K tokens relative to full-context KV caching, dramatically elevating procedural task execution in autonomous agents.

Researchers introduced ActionRoPE and Coordinate-Conditioned Noise Scheduling (CNS) in arXiv:2610.25861, solving spatial inconsistency in planar-camera game world models. In top-down and isometric gaming environments, naive feature-conditioned action mechanisms cause severe hallucination and revisit drift when players return to previously observed scenes. ActionRoPE directly converts action-induced camera displacements into world-coordinate rotary position embeddings (RoPE), ensuring identical physical world coordinates retain persistent token addresses across time. Benchmarks reveal that ActionRoPE cuts revisit drift by 66% over the strongest baseline while transferring across 1.3B to 33B parameter video backbones.

At PyTorch Conference North America on Oct 20, the CNCF Sandbox project llm-d released version 1.0. It turns inference engines such as vLLM and SGLang into one coordinated serving pool, and 1.0 adds multi-tenant priority and fairness scheduling (flow control), router high availability with zero-downtime rolling upgrades (DisaggregatedSet), a v1 configuration API, and agent-aware session routing with Responses API state. Founded in 2025 by Red Hat, IBM, Google Cloud, CoreWeave and NVIDIA, the project has 500+ contributors, and JetBrains, DigitalOcean and Mistral already run it in production.

On Oct 20 (UTC) DeepSeek open-sourced DeepInfer (Apache 2.0), the in-house inference stack that serves DeepSeek-V4.1-Flash at scale. The release covers offline inference with Attention DP/CP and MoE EP+EPLB parallelism, DSpark speculative decoding, SWA bounded replay, batch-invariant inference, deterministic sampling and structured outputs on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs. The KV cache system, prefill/decode disaggregation and load balancing are not included. DeepSeek also released deepseek-recipe (MIT), which converts Chat Completions, Messages and Responses requests and encodes them the same way the official API does.

Researchers introduced BlenderWorld (arXiv:2610.23986), the first comprehensive benchmark evaluating visual agents creating and editing 3D models from real-world references in Blender. Spanning 223 tasks across procedural geometry, Python scripting, material shaders, spatial layouts, and lighting, GPT-6 Astra leads the benchmark with a composite Score of 79.9 (and a 42.2% Perfect completion rate), followed by Opus-5 at 73.0. Open-weight flagship DeepSeek-V4.1-Flash claims 65.7, matching proprietary models while relying on active execution-and-inspection tool loops.

Scale AI Research introduced Sauron (arXiv:2610.17724), an open-source evidence-first agentic verification framework. Overcoming the severe failure modes of standard LLM-as-a-judge approaches—which hallucinate justifications and guess outcomes over long agent trajectories—Sauron enforces bounded, adaptive interrogation of agent traces with cryptographic-grade execution receipts. Evaluated on the PA100 benchmark across eight backbones, it delivers 0.94 precision and 0.97 recall at roughly 15% the cost of state-comparison baseline Archipelago, completing 10-episode grading 4.2x faster.

Researchers introduced MemPilot (arXiv:2610.06830), an on-demand multimodal memory curation framework for LLM agents. Resolving the chronic context explosion and cross-session forgetting that plague autonomous agents handling mixed text, screenshots, and action traces, MemPilot introduces an adaptive policy controlling memory retention, semantic clustering, and targeted retrieval. Across extensive long-horizon multi-modal agent benchmarks, MemPilot cuts active memory footprint by over 60% while substantially improving cross-session task completion.

The vLLM Project officially announced that PhysisServe's core physical-world serving techniques are being integrated into vLLM-Omni. Addressing the severe memory bottlenecks and latency spikes of multi-step video diffusion and physical AI generation, PhysisServe introduces cross-denoising-step pipeline parallelism across Transformer blocks, time-shared batching, and virtual-rank joint scheduling. Benchmarked on MiniMax H3 33B across 200 GPUs, it generates a chunk every 4.1 ms (~1,000 FPS without VAE decoding), while cutting single-GPU latency for Wan2.2 5B from 173 ms down to 98 ms.

A comprehensive empirical study evaluated native sub-agent concurrency across 2,124 runs on Claude Code, Codex, and Kimi Code (354 tasks, ~$20.8K API cost). On 100 standard SWE-bench Verified tasks, enabling sub-agents caused Claude Code's solve rate to collapse from 83% to 59% (with latency increasing from 17.5 to 27.8 minutes) and Kimi Code from 77% to 59%, while token consumption inflated by 1.41x to 3.31x. The study revealed severe failure modes including budget-exhausting fan-outs, orphaned child processes, and concurrent delivery race conditions.

In the latest independent Text-to-Speech (TTS) evaluations published by Artificial Analysis, Ant Group's open-source Ming-3.0-tts achieved a 93.2% pronunciation accuracy score, ranking #1 among all 31 evaluated models (outperforming commercial leader ElevenLabs v4 at 91.7%). Offering both a dense 0.5B parameter variant and a 7.9B mixture-of-experts (MoE) model, it delivers voice cloning and preset voices with a Controlled Voice Elo score of 1,069 (#1 among open-weight models), fully released under the permissive MIT license.

Independent AI evaluation authority Artificial Analysis launched the HealthBench and HealthBench Professional clinical reasoning and medical conversation leaderboards. Addressing the critical flaw where models artificially inflate medical scores through exhaustive, padded responses, the Professional benchmark implements rigorous length penalties. GPT-6 Astra (max effort) leads the world at 66.0%, with GPT-6.1 Sol at 65.1% and Claude Sonnet 5.5 at 63.0%. Claude Opus 5.5, despite holding a 71.4% raw score, collapses to 59.9% (#6) due to averaging ~5,900 characters per reply.

OpenAI officially announced breakthroughs in building and migrating software written in low-resource and legacy languages (COBOL, Fortran, Visual Basic, OCaml), introducing the MultiLangBench evaluation suite. Tackling the trillions of dollars in critical banking and enterprise mainframe infrastructure, the benchmark challenges models to synthesize a SQLite-compatible database engine across 20 distinct languages. GPT-6 Astra boosted average test pass rates from 26.8% to 87.4%, with COBOL soaring from 10.9% to 72.1%. In enterprise COBOL migration tasks, GPT-6.1 Sol achieved a 97.9% success rate while cutting inference costs by 39%.

Researchers from the University of Maryland, Google, and MBZUAI introduced BeliefWeaver (arXiv:2610.18082), an autonomous failure-recovery framework for long-horizon agents. Addressing cases where failures stem from root errors multiple steps earlier that local reflection cannot fix, BeliefWeaver maintains a persistent relational belief state across checkpoints and attempts, using an Implicit World Model (IWM) to evaluate hypotheses and gate repair execution. When distilled into a 9B model via on-policy distillation, it improves final benchmark scores by 9.8 percentage points while slashing teacher API costs by 54.8%.

Adobe Research and Brown University released OmniXRef (arXiv:2610.22744), a unified framework for multimodal multi-reference video editing. Overcoming the limitations of prior systems constrained to text prompts or single visual exemplars, OmniXRef utilizes a single-stream Diffusion Transformer with Mix-Patchify and Reference-Relative RoPE (R2-RoPE) to jointly control camera motion, subject content, dynamic motion, style, and visual effects (VFX). In camera trajectory evaluations, OmniXRef cuts rotation and translation errors by more than 5x compared to leading commercial engines.