Is Day 17 gone? Did Tibo fall asleep and forget?
Model intelligence feed
Model intelligence.
Follow the changes. Understand what comes next.
Editor’s pickOther providers
Valve's Answer to AI Slop and Clone Games on Steam: We Won't Police Developers' Tools
At Steam Indie Connect in Korea, Valve staff addressed AI slop and clone games head-on: no gatekeeping on AI tools, and the best defense against clones is community and steady updates.
Latest stories334
Newest published616263646566676869707172
Generate and Edit Images Directly in Copilot
GitHub’s Changelog dated October 28, 2026 says Copilot can generate and edit images with natural language. It is a public preview, not generally available. The selected reasoning model calls a dedicated image tool powered by GPT Image 2.5 Flare; results preview inline and save to the project. Available in VS Code, the Copilot app, and Copilot CLI. Org owners can disable the image model via Copilot model policies. Optional “Made with AI” watermark is off by default. Preview for Pro, Pro+, Business, and Enterprise; billed separately on image tokens toward AI credits. For reference only.
GitHub Copilot, GPT Image 2.5 FlareRead story
Shared Copilot Automations Roll Out for Teams
GitHub’s Changelog dated October 28, 2026 says Shared Copilot Automations are rolling out, not generally available by name. Teams combine prompts, triggers, and tools so Copilot runs in the cloud with only allowed tools. Shared ones are visible to repo readers and editable by writers; AI credits bill the repository owner; actions are attributed to Copilot. Org repos need the cloud-agent billing policy enabled. Rolling out to Copilot Pro, Pro+, Max, Business, and Enterprise; availability may vary. For reference only.
GitHub Copilot, Copilot Automations, Shared Copilot AutomationsRead story
GitHub Agentic Workflows Are Generally Available
GitHub’s Changelog dated October 28, 2026 says GitHub Agentic Workflows (GitHub Next, open source) is generally available. Automation is defined in Markdown and compiles to Actions YAML for issue triage, CI failure analysis, docs, and tests, reusing runners and policy. Safeguards include integrity filter, default read-only, sandboxed firewall, safe outputs, and threat detection. Free, Team, and Enterprise on github.com; default Copilot CLI engine needs a Copilot license. For reference only.
GitHub Agentic Workflows, GitHub Actions, GitHub Copilot CLIRead story
NVIDIA Opens Nemotron 3.5 Transcribe Weights
NVIDIA released open weights for Nemotron 3.5 Transcribe on October 28, 2026 (Hugging Face ID nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B). It is an offline speech/audio model for about 40 languages, with speaker-attributed transcription, translation, and audio Q&A; up to about one hour of audio and a 65,536-token context; license OpenMDW-1.1. NVIDIA AI cites a 5.33% Open ASR English figure; the HF card’s nine-set mean is 4.71%. Different sources and timing. Artificial Analysis cites 2.8% AA-WER (third-party, unretested here). For reference only.
Nemotron 3.5 Transcribe, NVIDIA-Nemotron-3.5-Transcribe-30B-A4B, NeMo SpeechRead story
GitHub Copilot Fusion Goes GA with Multi-Model Routing
GitHub’s Changelog dated October 28, 2026 says GitHub Copilot Fusion is generally available. Instead of picking a model for each task, selecting Copilot Fusion lets Copilot route and orchestrate which models it uses. Fusion Standard, Fusion High, and Fusion Max set effort. GitHub cites an Artificial Analysis Coding Agent Index score of 60 for Fusion Max. Paid Copilot plans get a limited-time 30% discount on model costs. Fusion is on all Copilot plans in VS Code, the Copilot app, Copilot CLI, and JetBrains IDEs. For reference only.
GitHub Copilot／Copilot Fusion／Fusion Standard／Fusion High／Fusion MaxRead story
OpenCode 1.19.5 Stops Failed TUI Worker Requests from Hanging
OpenCode published v1.19.5 on October 28, 2026 as a formal GitHub Release. The Bugfixes line says failed TUI worker requests now return an error instead of leaving the caller hanging (@Nowaker; PR #54936). Release notes list Bugfixes only—no new features. When a TUI worker handler fails, callers can see an error and exit rather than hang. npm package opencode-ai lists latest=1.19.5; a common path is npm i -g opencode-ai@latest. No pricing change appears with this cut—for reference only.
OpenCode / 1.19.5 / opencode-aiRead story
Mistral Puts Responses API in Public Preview for Agents
Mistral (October 28, 2026) puts the Responses API in public preview for building agents. Versus Chat Completions, each response returns the model’s answer, reasoning, and tool calls as separate items. Docs describe an OpenAI Responses format surface that is stateless—apps keep history and resend it in input. It works with all chat models hosted by Mistral, including Large 4. Preview capabilities include text, image/file input, structured output, streaming, function calling, reasoning, and input_tokens. No new Responses pricing was announced in this window—for reference only.
Responses API / Large 4 / mistral-large-4Read story
DiffuSpace Opens Clover-35B: Open-Weight MoE Diffusion LLM Around 35B
DiffuSpace (October 28, 2026) open-sources Clover-35B-A3B-Preview—an MoE diffusion LLM (~35B total / ~3B active), thinking mode, 128K context, Apache-2.0. Weights on Hugging Face, ModelScope, and GitHub; SGLang and llama.cpp supported. Lineage: HKU NLP Dream; HF base_model Qwen3.6-35B-A3B. Vendor-reported agent scores include SWE-bench Verified 73.6, Terminal-Bench 2.1 51.7, τ³ Overall 65.3; ~680 tok/s on one GB300 BF16 agent workload—for reference only. A decision model comes later. No OpenRouter clover/diffuspace ID yet; try-out is apply/chat, not an ungated public API.
Clover-35B／Clover-35B-A3B-PreviewRead story
Opera WebAI Early Preview: What Local Inference Saves, Which Devices Can't Run It
Opera (27 Oct 2026) opens WebAI early preview in Opera One, GX, and Neon: sites and extensions call local Prompt, Decision, and Semantic Embeddings APIs in-browser. No separate AI app, local server, or API key; page content used by local AI need not go to a cloud provider. Opt-in: Settings → Opera AI → Enable Web AI API. Supported: Win64 Intel/AMD and Apple silicon Macs; not Intel Mac, Linux, or Windows on Arm. Six demo extensions are samples, not built-in. Proposed web standard—not a finished universal API. Speed and CJK quality not tested here. For reference only.
Opera One／Opera GX／Opera Neon／WebAIRead story
Liquid AI Releases LFM3: Open-Weight Multimodal Family Around 2B
Liquid AI (October 27, 2026) announces LFM3, an open-weight multimodal family around 2B on one shared backbone—text, vision, audio, omni, and decision (d1). Weights are on Hugging Face, Liquid Playground, and the Liquid AI API. LFM3-1.7B is about 1.73B parameters with 256K context; Liquid reports a tri-hybrid architecture and about 22T pre-training tokens. OpenRouter is a named day-0 partner, but public OpenRouter LFM3 model IDs were not listed yet—today's entry points are HF / Playground / Liquid API. Blog benchmarks are vendor-reported. For reference only.
LFM3-1.7B／LFM3-VL／LFM3-Audio／LFM3-Omni／d1Read story
GitHub Copilot Adds Server-to-Server Auth in Public Preview
GitHub’s Changelog dated October 27, 2026 says GitHub Copilot supports server-to-server authentication for GitHub Apps installed on organizations, in public preview. Services and agents built with the Copilot SDK can request on behalf of an organization without signed-in user credentials. Apps need the read-only Copilot Requests organization permission and a short-lived installation access token. Usage is billed to the org under its model policies, budgets, and spending controls. Tokens expire after one hour. Details follow GitHub Changelog and docs. For reference only.
Copilot／Copilot SDK／GitHub AppsRead story
OpenBMB Open-Sources MiniCPM-OCR: Parser Plus Result Checker
OpenBMB (News [2026.10.28]) open-sources MiniCPM-OCR: a parser-plus-checker suite, not Baidu Unlimited OCR 2 and not a MiniCPM-V 4.7 release. MiniCPM-OCR 1.0 (~1.3B, based on MiniCPM-V 4.6) handles document OCR; MiniCPM-OCR Jev (~2B) scores candidate results without parsing text and can follow other parsers. Vendor-reported figures include Real5-OmniDocBench 93.68, OCRBench v2 54.7, and JevOCRBench 93.6; DSpark claims cite ~55% efficiency and ~1.2 pages/s. Apache-2.0 weights on Hugging Face and GitHub. For reference only.
MiniCPM-OCR 1.0／MiniCPM-OCR JevRead story
From the Webmaster
Personal experiences, thoughts and notes.