Latest news, beside the leaderboard

On Oct 7 OpenAI began rolling GPT-6 out globally in ChatGPT together with Intelligent UI: the model decides when to answer with diagrams, interactive charts, forms, tappable buttons or small in-chat tools (calculators, games, bill splitters) instead of plain text. Plus, Pro, Business and Enterprise get it first on GPT-6 Sol; Go and free users follow on Oct 8 on GPT-6 Luna. OpenAI says GPT-6 can stream partial answers while still thinking, cutting wait times by 44%.

On Oct 7 Anthropic shipped Claude Haiku 5.5 (claude-haiku-5-5), completing the Claude 5.5 family. It targets high-volume, cost-sensitive work such as summaries, compaction, classification and coding subagents, and Anthropic calls it its fastest model at standard speed. Prompts up to 100K cost $0.10 input / $0.50 output per million tokens, and it is the first Haiku with effort levels. It scores 72.4% on the OSWorld 2.1 offline subset (Haiku 4.5: 15.7%) and 39.2% on Terminal-Bench 4.0. Sonnet 5.5 cache reads were cut 50% the same day, and Max/Team plans get monthly API credits.

On 2026-10-06 the FCC granted-in-part SpaceX’s D2D Constellation (S00735) under DA-26-1078: up to 15,000 sats at 326–335 km, with a §25.125 lease waiver plus PCS G and EchoStar-assigned bands. Milestone: 50% by 2032-10-07. ~150 Mbps/user is a company target, not an FCC measurement.

REA (Reverse Engineer Anything, MIT) wraps Hopper, Ghidra and IDA Pro into an MCP server and CLI for coding agents: 41 native inspection tools and 14 investigation workflows across Mach-O/ELF/PE, .NET, Electron, websites and Android APKs, all running locally. Version 4.1.0 (Oct 6) adds headless-JADX APK analysis and Binwalk/Unblob firmware analysis. The repo has 12,962 stars and 1,375 forks.

On Oct 7 Perplexity released pplx-embed-v2-late, multimodal late-interaction (ColBERT) retrievers in 0.6B (340M active) and 9B (7.4B active) sizes, built on Qwen3.5 with bidirectional attention. Each token gets a 128-dim vector scored by MaxSim, covering text, images and rendered PDFs or slides without OCR. Both sizes share one embedding space, so the 0.6B model can query a 9B-built index. Public ViDoRe v3 image nDCG@10 is 62.3% / 65.2%; MIT-licensed weights are on Hugging Face.

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.

On 2026-10-07 Elon Musk said Grok @Bot / @SpaceX will use the best backend model per task—naming Claude Opus 5.5, Midjourney, Suno and other leading APIs. Product context: [Introducing Grok Bot](https://x.ai/news/introducing-grok-bot) and [docs](https://docs.x.ai/grok-bot/overview). Routing rules, pricing, and timeline are still unspecified in official docs.

LLM systems handle evidence retrieval via two disjoint paradigms: external Retrieval-Augmented Generation (RAG) pipelines relying on auxiliary retriever-reranker stacks, and brute-force long-context inference suffering from quadratic attention overhead. NVIDIA and Technion introduce UNREAL (UNifying REtrieval And Long-Context with a Single Model, arXiv:2610.08463), a model-native framework unifying corpus retrieval and long-context filtering within a single frozen LLM. By deriving chunk representations and search queries directly from internal transformer activations, UNREAL introduces fewer than 500K parameters while keeping the backbone frozen. On a 3B-token, 21M-chunk Wikipedia corpus, UNREAL surpasses competitive retriever-reranker systems, lifting HotpotQA recall from 49.1% to 73.2% and 2WikiMultiHopQA from 31.7% to 60.1%. When applied to long-context sequences, UNREAL dynamically purges irrelevant distractors prior to generation, boosting 128K NoLiMa accuracy from 1.0% to 24.83% and 256K LV-Eval F1 to 54.66% while substantially curbing FLOPs and time-to-first-token (TTFT) from 32K context onward.

LLM agents operating in novel API or terminal environments frequently repeat identical errors due to an absence of procedural memory. Existing memory bootstrapping approaches rely heavily on curated prompt playbooks or oracle verifiers that require extensive domain foresight. Illuin Technology releases DAEDALUS (arXiv:2610.08048), an autonomous dual-agent framework bootstrapping reusable memory entirely from self-generated practice without human oracles. DAEDALUS pairs an Explorer agent—which probes the execution environment to formulate curriculum-style tasks—with a Solver agent. When the solver stumbles, it extracts candidate heuristics that are committed into a persistent memory bank only after proving their efficacy across repeated verification trials. Across AppWorld, tau^2-bench, and AutomationBench, DAEDALUS boosts average success rates by up to 15.9 percentage points and raises Pass^5 rates by 2.2x over memory-less baselines, with generated procedural memories transferring seamlessly across heterogeneous foundation model families.

Deploying performant coding and OS agents on cost-efficient small language models (SLMs) is essential for enterprise scalability. In modern agent architectures, models operate within an execution harness that maintains workspace context, API state, and terminal feedback. Conventional distillation forces the student model to mimic the teacher's holistic sequence output, conflating harness-maintained facts with model reasoning and overloading the student's limited capacity. KAIST AI introduces Harness-Aware Distillation (HAD, arXiv:2610.02858). HAD decouples what the harness manages from the unique reasoning additions of the teacher model. By combining on-policy distillation with harness-conditioned action preference contrast and strict execution-record validity checks, HAD requires zero reward models or oracle success annotations. Across multiple long-horizon agent benchmarks, HAD-trained SLMs enter significantly fewer unproductive query loops and demonstrate substantially higher autonomous error-recovery rates.

On Oct 7 LiteLLM open-sourced Moyai (BerriAI/moyai), the self-hosted background coding agent its team runs daily: tasks from Slack or the browser run in isolated cloud workspaces that edit code, run tests and open PRs. Harnesses (Claude Agent SDK, Codex, Hermes, OpenCode, Deep Agents, Tool Loop) are switchable per session and inference routes through LiteLLM's 100+ providers. LiteLLM says replacing Devin cut a 31-day bill from $101,872 to about $21,700.

Tool-using agents execute consequential modifications to external systems, yet nominal task success frequently conceals unverified actions executed without prior evidence. Researchers from the National University of Singapore (NUS) present a rigorous empirical study on the breakdowns along the evidence-to-action chain (arXiv:2610.07753). Evaluating ten model-harness configurations reveals a striking paradox: high static action assessment accuracy routinely masks brittle interactive execution. Failures overwhelmingly originate prior to execution: agents terminate prematurely before collecting requisite evidence, or initiate state-altering actions before foundational proofs are established. The authors formulate SafeActBench, spanning 656 rigorous cases across six operational domains and five protocols progressing from static judgment to complex multi-action workflows. Supported by a provenance-bound Evidence Ledger and deterministic trajectory evaluator, findings demonstrate that while single-step execution is stable once evidence is verified, multi-action workflows frequently collapse under unresolved prerequisites and truncated state transitions.

Continual skill evolution allows LLM agents to autonomously accumulate, refine, and reuse procedural knowledge across multi-turn interactions without updating model parameters. However, prevailing experience-driven reflection mechanisms suffer from two fundamental weaknesses: they decouple synthesized guidance from concrete behavioral traces, and they rely on monolithic global validation that blindly discards valid local corrections whenever an entire workflow fails. Researchers from Tsinghua University and partner institutions introduce EVISKILL (arXiv:2610.05030), an evidence-grounded framework anchoring procedural adaptation to deterministic execution traces. EVISKILL structures raw observations into Replayable Evidence Cards and synthesizes skill edits with explicit contextual lineage. A targeted replay engine verifies edits through isolated re-execution before cross-epoch refinement. Across three interactive benchmarks spanning six LLM backbones, EVISKILL consistently outperforms state-of-the-art reflective baselines, demonstrating durable procedural accumulation.

Agentic reinforcement learning (RL) disaggregates centralized policy training from large-scale interactive environment rollouts, necessitating frequent cross-cluster policy weight synchronization (refit). Transferring a full 1T-parameter checkpoint across cloud regions currently consumes 87.5 minutes, idling massive rollout compute clusters. NVIDIA researchers introduce NeMo-DCR (Delta-Compressed Refit, arXiv:2610.08430), a bit-exact synchronization architecture that transmits only model deltas while guaranteeing receivers reconstruct identical parameter bits. Exploiting empirical measurements showing only ~1% of BF16 weights alter value bits per step, NeMo-DCR maps training shard changes into canonical coordinates via fixed affine projections and utilizes compressible XOR masks to guarantee exact bit-level parity. Streaming deltas over relay trees and committing changes via atomic joint commits, NeMo-DCR achieves 12x to 40x speedups for models spanning 30B to 1T parameters, slashing 1T cross-region refit latencies from 87.5 minutes to 150 seconds.

At Team '26 Europe, Atlassian introduced AMP, a framework giving agents an assigned identity, scoped authority, shared Teamwork Graph context and reviewable output, and made a rebuilt Atlassian MCP generally available with 220+ tools, OAuth 2.1/PKCE and tiered read/write/destructive scopes.

On 2026-10-06 Google officially launched Nano Banana 2.1 and marked it GA in the Gemini API changelog as gemini-nano-banana-2.1. It is based on Gemini 3.6 Flash and is rolling out across the Gemini app, AI Studio, Search AI Mode, Flow, Stitch, Google Ads and Gemini Enterprise Agent Platform. Image output is priced at $30 per 1M tokens (Nano Banana 2: $60), which works out to $0.0336 per 1K image and $0.0504 per 2K image, about half the old price. The model card's side-by-side human eval puts text-to-image Elo at 1050, ahead of Nano Banana 2 (990) and Nano Banana Pro (935). gemini-3.1-flash-image is now deprecated, with no shutdown date announced. This confirms the silent Flow rollout we reported earlier.

On 2026-10-06 GitHub made stacked pull requests generally available on all github.com plans, with GitHub Enterprise Server support coming in a later release. Since the public preview, repos using stacks have merged 9% more code than their peers, and more than two-thirds of the top 1% of repos now use them, with a 5% faster time-to-merge. New in GA: approvals are kept for unchanged code after a rebase, replacement commits are signed automatically, a whole stack enters the merge queue as one merge group, stacks retarget automatically when their base branch is deleted, and stack auto-merge is rolling out over the coming weeks. The gh-stack CLI extension (v0.2.0) now supports Git worktrees, and `gh skill install github/gh-stack` teaches AI coding agents how to split large changes into stacked PRs.

On 2026-10-06 Anthropic merged Project Glasswing and the original Cyber Verification Program into one three-tier program: Defense Access (SOC, incident response, malware reversing, vuln validation), Red Team Access (authorized pentesting) and Specialized Access (safety-critical systems such as power grids, flight systems and telecom, vetted with the US government). All tiers get Claude Opus 5.5, Sonnet 5.5 and Mythos 5.1 with reduced cyber blocking. On CyScenarioBench, GA models were blocked on the first prompt of every task, Defense tier blocked 46/50 trials, and Red Team tier had zero blocks and completed 34/50 (matching the 67.6% no-safeguard rate). Glasswing partners reported at least 129,000 verified vulnerabilities from April to July 2026.

FeSens/openTPU (Apache-2.0) hit the Hacker News front page (250+ points, 300+ comments). Reusing the AI-driven method from auto-arch-tournament (AI-designed RISC-V cores), the author had AI agents iterate on an inference accelerator; one repo holds SystemVerilog RTL, the ISA, a bit-exact simulator, a kernel language/compiler and a profiler. On a Kintex-7 xc7k480t PCIe card it runs Qwen3, Qwen3.5, Gemma 4, Phi-4-mini and others with real weights, token-for-token identical to the simulator; LFM2.5-230M decodes at 85.8 tok/s in 4-bit and Qwen3.5-35B-A3B reaches 3.95 tok/s with host-streamed experts.

Large Language Model (LLM) coding agents equipped with bash terminals frequently collapse when deployed across massive enterprise codebases. Centralized agents struggle to reconstruct project states fragmented across multi-language source trees, build configs, and test suites, precipitating context window exhaustion and catastrophic semantic drift. Researchers from UIUC introduce Dev-Primitives and the HERMES harness engineering framework (arXiv:2610.07832). Dev-Primitives transform passive software components into active agentic nodes by pairing each repository artifact with a resident LLM endowed with natural-language reasoning, inter-component communication channels, and localized self-modification primitives. Coordinated via a dependency-aware dynamic activation scheduler and execution-guided fault localization, HERMES boosts task performance across four standard software engineering benchmarks by 12.4% over baseline harnesses. Crucially, powering HERMES with lightweight Qwen3-8B Dev-Primitives brings performance within 4.5% of homogeneous GPT-5.6 Sol configurations while slashing inference token overhead by 26.2% on Terminal-Bench 4.0.