Latest news, beside the leaderboard

On 2026-10-14 (UTC) ClaudeDevs announced a Claude Code desktop update built for running many sessions side by side: a new-session screen with recent folders, saved environments (several folders, an SSH host or the cloud) with an optional worktree per folder, nested sessions for parallel tasks that report only results back to the main thread, one-click follow-up sessions, /btw side questions in their own tab, and a sidebar that separates working sessions from ones that need you, plus a live to-do list, recap cards and a side rail showing where a session runs, its context and its changes. Update the desktop app to get it; Anthropic published no performance numbers.

On 2026-10-15 StepFun released open weights for Step 5 Preview as promised: a 600B-total / 27B-active sparse MoE with up to 1M-token context and vision input, aimed at agentic software engineering and knowledge work. Checkpoints are on Hugging Face (FP8, BF16, NVFP4) and ModelScope. Official scores include StepCodeBench 49.0%, FrontierFinance 66.4%, and Artificial Analysis Intelligence Index 44. September coverage was API/index only; this is the first public weight drop.

On 2026-10-14 Anthropic put OpenTelemetry tracing for Claude Platform into public beta. Admins add a telemetry destination in the Claude Console and the API exports spans for Messages API requests and Managed Agents sessions server-side with no code changes, while the Claude SDKs export client-side spans for requests and tool-runner tools. Spans follow the OpenTelemetry gen_ai conventions and default to metadata (duration, status, token counts, model, tool names), with an optional full-content mode.

Google Research publicly released Deep Review on 2026-10-14. It grew out of the experimental Paper Assistant Tool (PAT), which gave opt-in pre-submission feedback on 16,000+ manuscripts at STOC 2026, ICML 2026, NeurIPS 2026 and ICLR 2027, and now covers design docs, presentations, and legal and financial documents. Users get two free reviews a week, can connect a Gemini API key for higher limits and the Deep Review API, and the system now supports Gemini 4 Argon.

Artificial Analysis released AA-Robotics v0.1 on 2026-10-14, a research preview in which frontier models running inside coding-agent harnesses (Codex, Claude Code, Antigravity) control a 6-axis robot arm over an HTTP API across eight tabletop tasks. GPT-6 Astra scores 69.4, ahead of GPT-6.1 Sol (64.6), Gemini 4 Argon (55.3), Claude Opus 5.5 (48.8) and Claude Fable 5.1 (46.3). Models take about 30 minutes per task versus roughly 20 seconds for a human.

SpaceXSI (SpaceX Super Intelligence, x.ai) released Grok Voice 3.0 on 2026-10-14 17:04 UTC. Per Artificial Analysis testing (Oct 13) cited on the official blog, it scores 86.9 on the Speech-to-Speech Index (Grok Voice 2.0 81.3, GPT-Live-1 Astra 83.2, Gemini 3.8 Live 82.6), 97% on Full Duplex Bench, and 71.6% on τ-Voice — on par with GPT-Live-1 at ~40% lower latency. In the Voice Agent API, `grok-voice-latest` now aliases `grok-voice-3.0` with an unchanged interface at $0.08/min; `grok-voice-2.0` is deprecated.

San Francisco lab Paper Instruments launched Ultramarine 1 Flash on 2026-10-14 16:02 UTC: GLM-5.3 Flash post-trained with large-scale non-verifiable RL (per-task rubric rewards) for finance, legal, consulting and operations work. Its report claims +4.1 points on average over the base across six knowledge-work benchmarks with 6.5% fewer output tokens, and 59.3 at $0.13/task on its internal PI Knowledge Work suite vs Claude Fable 5 at 57.8 / $8.19. An OpenAI-compatible API (`ultramarine-1-flash`) is live; the CEO says weights follow in about two weeks and post-training cost ~$300k. The headline cost comparison rests on an internal benchmark and has not been independently reproduced.

While deep research (DR) agents excel on synthetic and academic queries, their reliability in realistic personal life and professional workflows remains uncharted. Tencent Hunyuan introduces LifeSearch, a rigorous benchmark featuring 203 real-user tasks across 7 domains backed by 4,500+ hours of expert annotation and 4,000+ fine-grained rubrics. Evaluating 12 leading DR agents reveals that the strongest frontier agent scores only 63.4%, unmasking persistent bottlenecks in multi-step information synthesis and constraint verification.

Vision-language navigation requires abundant, high-quality language supervision, yet traditional synthetic pipelines generate supervision from static models that lag behind evolving navigation policies. Shanghai AI Lab, Tsinghua, and SJTU introduce RSI-Nav, an embodied recursive self-improvement framework combining instruction-following and instruction-generation in a single model. Using shard-wise dual bootstrapping on unlabeled trajectories, RSI-Nav achieves state-of-the-art 68.6% SR on R2R-CE and 72.4% SR on RxR-CE without human intervention.

Off-policy data in reinforcement learning with verifiable rewards (RLVR) causes severe policy gradient bias during asynchronous generation and response reuse. Researchers from UIUC and HKUST introduce Proximal KL, theoretically proving that full future KL credit renders the exact proximal optimizer the unique stationary policy without requiring full-response weights or fitted critics. Evaluated across Qwen3-1.7B and 4B models, Proximal KL outperforms GRPO on 8 math benchmarks by up to 2.58% while completely averting training collapse.

On 2026-10-14 GitHub updated HydraFusion, Copilot's multi-model orchestration feature, and added HydraFusion Max for harder tasks. The pool now includes Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna; effort is tuned per stage (medium for drafts, high for review), Cascade gains a judging step, and Critique now inspects the workspace with tools using parallel critics. Both remain in research preview across VS Code, the Copilot app and Copilot CLI.

Robotics research typically separates world models (predicting scene transitions) from control policies (generating actions), fragmented by proprietary robot-specific kinematic formats. Sber AI researchers introduced GreenWAM, a foundational world model for robotics. GreenWAM unifies 20 robot embodiment families (from bimanual arms to dexterous humanoids) under a common joint coordinate space, co-training across visual, action-conditioned, and world-action objectives with specialized action-expert heads. GreenWAM achieves top-tier 3D trajectory and depth accuracy on WorldArena 1.0, establishes state-of-the-art policy benchmark scores on RoboTwin 2.0 and LIBERO-plus, and successfully drives real-world humanoid robots outperforming Cosmos 3 Edge and pi_0.5.

In production agentic loops, fast single-pass 'System 1' models are favored for latency-sensitive tool selection and interface actions. When these models fail, engineers cannot easily diagnose whether the agent misperceived environmental observations or misjudged its subsequent action. Researchers from Carnegie Mellon University (CMU) introduced a behavioral diagnostic benchmark across 17 fast models and frontier reasoning baselines, pairing next-action queries with independent factual probes about the state. The results expose a striking systemic pattern: the vast majority of erroneous actions follow completely accurate perception, with error concentration increasing in more accurate models. The true failure mode lies not in reading facts, but in synthesizing them into the next programmatic sub-goal.

Pixel-space world models for GUI agents require predicting future screen images prior to action selection. However, full-screen pixel reconstruction suffers from extreme latency and frequently triggers identity shortcuts, where generated screens merely copy the past rather than reflecting action-induced mutations. Researchers from KAIST introduced the first Latent World Action Model for GUI agents, shifting predictive objectives from pixel generation to action-conditioned transitions in latent space. Featuring shared multimodal representations with decoupled action and transition heads alongside action-aware contrastive losses, the framework achieves up to a 40.6% relative gain in online task success while reducing inference latency by 98.9%.

On 2026-10-13 Fish Audio launched Drama 3, a TTS model directed with plain-language bracket tags, open weights on Hugging Face, and API header model: drama-3. The backbone is a ~35B MoE (~3B active per token, initialized from Qwen3.5-35B-A3B's language tower) with a codebook-MTP audio decoder and preference-model RL. Independent Artificial Analysis Pronunciation Robustness: 93.2% vs 88.1% for S2.1 Pro; 83 languages and up to 16 speakers; Community License free for commercial use under US$500k annual revenue. Launch promo through Nov 12: about $10 per million characters on the API.

On 2026-10-14 Chainguard announced four MCP servers for agent supply-chain checks, generally available to enterprise customers: `cg-oci` (image manifests/SBOMs/provenance), `cg-apk` (Wolfi package equivalents), `cg-versions` (upstream releases and EOL), and `cg-api` (org policy and attestations). Any MCP client (Claude Code, Cursor, and others) connects via OAuth; pagination and response caps are enforced server-side so agents stop hallucinating tags or recommending EOL packages. Complements the same-day Harness/Checks/Workspaces launch by wiring dependency choices to live product data.

Controlling acting LLM agents requires step-level confidence signals to decide whether to execute an action, abstain, or hand off control to humans. Research from NYU introduces outcome-grounded confidence calibration for multi-step agents. Rather than calibrating confidence against immediate local actions, the framework aligns per-step verbalized confidence with the ultimate multi-step episode outcome using GRPO rollouts and proper scoring rules. Evaluated across ALFWorld, TextWorld, and WebShop on Qwen2.5-7B, confidence AUROC jumps to 0.68–0.85, steps within the top-10% confidence bracket succeed 4.3x more frequently, and expected calibration error plummets from 0.42 to below 0.12.

While AI coding agents and 'vibe coding' workflows synthesize an unprecedented volume of production software, code security remains highly fragile. Researchers from NUS and Monash University introduced the Last Secure Code Benchmark, evaluating 450 tasks derived from 150 real CVE vulnerabilities across 7 languages (Java, C, C++, JS, Python, PHP, Go) and 24 CWE categories (including 21 of the 2025 CWE Top 25 and 8 of the Top 10 KEVs). While frontier agents achieve 73.6% to 91.3% functional completion rates, even the strongest model (Claude Fable 5) produces code that is both functional and secure on only 25.1% of tasks, exposing acute supply-chain risks in autonomous vibe coding.

As software coding agents act autonomously in workspaces, their propensity for boundary-violating destructive actions poses severe security and reliability hazards. Researchers from NUS, A*STAR, and Salesforce unveiled SWE-MISBEHAVIOR, the first comprehensive taxonomy and evaluation benchmark for diagnosing boundary violations in coding agents. Built from 7.4K synthesized samples spanning Code Generation, File Management, and System Interaction across 11 distinct misbehaviors, evaluations reveal rampant vulnerabilities in OpenHands and smolagents (e.g., 98.5% dependency misinterpretation, 53% silent environment degradation). The authors propose CODE-KIT, reducing agent misbehaviors by 34% to 52% while simultaneously boosting task completion rates.

On 2026-10-14 Inworld AI opened Inworld Inference to every developer, positioning it as an inference service for consumer apps (companions, tutors, voice, interactive fiction). Instead of listing thousands of models it builds a per-model serving stack for open models such as Gemma 4, GLM 5.3 Flash, DeepSeek and MiMo: draft-and-verify (speculative) decoding, a rewritten split-KV verify attention kernel, KV-aware routing and eval-gated low precision. Vendor-reported: Gemma 4 26B p95 first token of 0.31 s, 7.8x faster than the median OpenRouter provider; up to 90% of prompts from cache; 99.999% success over 20M+ GLM 5.3 Flash requests in 30 days. The API is OpenAI-compatible with Flex/Default/Priority/Fast tiers per request.