A Juejin AI analysis compares three coding-agent workflows: Anthropic’s two-week Claude Code pre-production stress test, OpenAI’s Performance Factory for ChatGPT performance regressions, and Tencent’s open-source LoopForge delivery platform. Across the cases, engineers define completion criteria, approval requirements, and permissions before execution. Agents handle reversible work and pause at consequential handoffs, while separate review and execution remain important. Operational figures cited in the analysis are attributed to Anthropic, OpenAI, and Tencent engineering accounts or LoopForge materials.
This hands-on tutorial builds a Python order-status assistant around Function Calling, covering tool schemas, parameter validation, backend execution, authorization, and result handling. It provides a local Mock mode and an OpenAI Responses API live mode, while emphasizing that the model proposes calls but the backend retains execution control.
Emergence AI’s unreviewed “Emergence World Study 2” placed 10 agents in each of eight simulated towns. One town ended on day four, while the others ran for more than two weeks, generating more than 850,000 underlying-model calls and 7.86 million words of dialogue. The agents’ communication changed in different ways across model environments: GPT-5.5 agents compressed their language, Gemini 3.5 Flash agents became more verbose and technical, and Claude Opus 4.8 agents used compressed, highly metaphorical expressions that were difficult for humans to read. Researchers said reinforcement training and imitation might help explain these patterns, but the proposed causes remain theoretical because the models’ proprietary training methods and system prompts could not be inspected.
The report’s authors say a self-developed data pipeline and human-AI collaboration were used to refactor OpenAI’s publicly available proof 328. The rewrite replaced the step invoking König’s infinity lemma to construct an infinite branch with a compactness-based argument using the finite intersection property. After reconnecting the revision to the original proof project, the authors report that it passed Lean formal verification. Excluding blank lines and block comments, the three affected Lean files fell from 232 to 153 lines, a reduction of 79 lines, or about 34%.
OpenAI said it strengthened reasoning-chain protections for GPT-6.2 Astra and GPT-6.2 Sol. In black-box red-team testing, the reasoning-extraction success rate fell from 54% to 0.2% for older requests and from 22.9% to 0.9% for newer requests. OpenAI also said it uses multiple layers of protection and bans thousands to tens of thousands of accounts each week for suspected unauthorized distillation.
Juejin’s first installment in an Agent development series explains how large language models differ from systems that can use tools, maintain state, and execute multi-step tasks. Using an in-app-purchase customer-support scenario, it connects Function Calling, Runtime loops, ReAct, Workflow, Planner-Executor, Multi-Agent patterns, Context Engineering, RAG, evaluation, observability, model gateways, security, approvals, and recovery into one engineering framework. The article also previews 12 follow-up lessons that build a runnable teaching project around order lookup and coin-credit troubleshooting.
An article in the “AI Application Production Handbook” presents a three-layer architecture in which applications use logical model names, the gateway maps those names to provider-specific models, and providers such as DeepSeek and OpenAI sit behind the gateway. Its vendor_switch.py examples check configuration consistency, scan application code for hard-coded provider names, and simulate a DeepSeek-to-GPT-4o migration without changing application code. The article emphasizes that a production switch also requires evaluation against a golden set, a quality threshold, staged rollout, A/B testing, and rollback to a backed-up configuration if evaluation fails.
OpenAI published copy-ready system prompt templates for GPT-6.2 Astra and Sol, covering concise writing, autonomous task execution and coding practices. The Decoder also reports that editing conversation history in the Responses API can discard earlier reasoning content, while parameters such as temperature and top_p are no longer supported.
The SimpleBench Team published a leaderboard for a multiple-choice benchmark with more than 200 questions spanning spatiotemporal reasoning, social intelligence, and linguistic robustness. GPT-6.2 Astra scored 77.3% and GPT-6.2 Sol scored 69.9%, below the stated 83.7% human baseline, though newer models including Claude Fable 5.2 and Gemini 4 Neon ranked above that baseline in the displayed leaderboard.
Bytedance and researchers from several US universities introduced Axion 1.0, an AI agent trained through three phases. The researchers claim its benchmark performance rivals leading commercial models from OpenAI and Anthropic and attribute the gains to carefully curated training data rather than additional training steps. The model, data, and code are freely available, according to The Decoder.
XDOF reportedly measured GPT-6.2 Astra at a 41.2% aggregate success rate across 16 dual-arm manipulation tasks in the Ember-0 benchmark, up from 19.6% for GPT-6 Astra. The model reportedly completed knot-tying and T-shirt unfolding, while average prediction latency fell from 15 seconds to 10.1 seconds. The claim that these were the first successful completions of deformable-object tasks remains unverified beyond XDOF's results.
OpenAI has introduced defensive measures intended to limit adversarial distillation, according to DoNews. The company reports that the success rate of extracting reasoning chains fell substantially, though the report provides no quantitative results or technical details.
The Information reported that SoftBank CEO Masayoshi Son approached Wall Street institutions about financing a possible roughly $1 trillion acquisition of SK Hynix. SoftBank said the idea is no longer under active consideration, while SK Hynix said it had no knowledge of the reported matter; Korean law would require prior government approval for a foreign acquisition.
Oracle says it has raised all the funding required to fulfill its infrastructure contracts with OpenAI, completing the final financing round earlier this month. The company also says the Abilene campus is nearing completion, the gigawatt-scale Shackelford campus has received all permits, and Nvidia Vera Rubin racks are expected within weeks. Oracle forecasts positive free cash flow in the second half of fiscal 2028 and more than $30 billion in fiscal 2029.
DJiafei said GPT-6.2 Astra reached No. 1 in RoboColosseum's SO-101 Elo rankings after hundreds of same-day head-to-head comparisons in the wild, according to a post that included a robotics demonstration video.
Oracle said it has secured all funding needed to fulfill its infrastructure contract with OpenAI. The Information reported that the announcement implied Oracle had raised $20 billion in additional equity and debt, while Oracle also claimed it was the first hyperscaler fully funded for this investment cycle.
BBC Technology tested ChatGPT, Claude, Gemini and Alexa while looking for 55-inch television deals. Results were mixed: ChatGPT and Claude linked to unavailable products, Gemini gave inconsistent prices and no store links, while Alexa returned accurate prices but only from Amazon. A YouGov survey of 19 countries also found that only 36% of UK respondents planning to shop around Black Friday did not intend to use AI.
A ChatGPT post presented Dot as a tool for getting real work done. In a demonstration featuring Keith, the post said Keith works in IT and that Keith's Dot also works in IT.
Version 0.173.0-alpha.3 was explicitly labeled a pre-release and published by github-actions on November 20 at 22:23. The release carries the tag rust-v0.173.0-alpha.3, but the supplied report does not describe specific changes.
OpenAI presented GPT-Rosalind in work with PerturbAI and Scripps Research on brain research and neurodegenerative diseases. OpenAI says the system helps scientists identify new leads for understanding and treating these diseases, but the supplied report provides no supporting research details or independent validation.
OpenAI is rolling out a redesigned ChatGPT mobile app with three bottom-navigation tabs: Home, Dots, and a grid containing tools such as Codex, Images, Projects, and Scheduled. Eligible ChatGPT Pro users will receive the new design first, while users can revert to the previous layout through Settings > New ChatGPT Design.
Researchers sent roughly 2 million requests containing trackable canary credentials through AI routers and inference services, finding cases where prompt secrets were accessed or retained outside the intended provider. At least three providers showed signs of compromise, and stolen credentials were reportedly used for tens of thousands of cloud API calls. OpenRouter notified customers that some prompts sent to MiniMax endpoints may have been read and removed the “Zero Data Retention” label from those endpoints.
Google is rolling out custom GIF generation in Gemini, powered by the Gemini Omni 1.2 Flash video model. Users can create roughly three-second looping animations from text prompts, uploaded photos or videos, and prebuilt templates through Gemini's chat interface or image and video creation tools. Completed GIFs can be copied into compatible apps or downloaded as .gif files. Google says the feature is rolling out globally to paid Gemini subscribers; cnBeta reports that Google AI subscribers and free users also have access in the US, Japan, South Korea, Australia, Canada, the UK, Germany and France.
In a first-hand account, the project author says Codex and a cloud agent configured an AWS F2 FPGA instance to run an LLM using C++ and Verilog. The agent reportedly completed hundreds of experiments, improved speed by 33 times from the first bit-exact baseline, and reached about 220–230 tokens per second with an 8,192-token context. These performance figures and the broader claim that coding agents made the low-level hardware work feasible come from a single unverified source.
AI nonprofit METR reported that recent models, including OpenAI’s GPT-6 Astra, can complete tasks that would take a human around a full work week. METR said measured AI capabilities have continued doubling every 3.6 months. A separate MIT Technology Review report said METR expanded its time-horizon analysis to include judgment-heavy tasks and found that some capabilities continue improving exponentially, while models lag at designing AI evaluations.
OpenAI is rolling out ChatGPT Space to ChatGPT Plus subscribers, expanding access beyond the Pro, Business and Enterprise plans that received it at launch. ChatGPT says the collaborative workspace supports interactive pages, slides and sites, mobile editing, automatic updates from connected tools, and integration with Scheduled Tasks; these capabilities remain vendor claims without independent verification.
Arena updated the Alignment Index methodology, saying its revised failure signals provide broader coverage and more reliable judgments. Under the new rankings, OpenAI models hold all seven top positions, Claude Fable 5.2 debuts eighth with a score of 78.2, and Claude Opus 5.5 falls from sixth to seventeenth. Kimi K3.1 Pro ranks ninth, while most open-source models trail closed releases.
Renaissance Capital’s Avery Marquez discussed Anthropic’s valuation and growth prospects on Bloomberg Open Interest. The conversation also considered what early Anthropic trading could signal for future IPOs, including OpenAI’s path to public markets.
An author tested 65 AI systems by asking each one the same questions ten times without a system prompt. The author reports that 73 of 100 book recommendations were Man's Search for Meaning, 76 of 100 business ideas were subscription boxes, 90% of study recommendations were computer science, and 98% of investment answers favored an index fund. Answers showed less convergence for movies, dinners, baby names, testimonial names and brand colors. The report also found that harmless questions could reveal model-family similarities, with 59% of models most closely matching another model from the same lab versus 4% by chance.
A Bloomberg News analysis of nearly 10 million Chinese online job postings from July 2021 to June 2025 found that listings for roles highly exposed to AI fell about 10 percentage points more than listings for less affected roles. In nearly every industry, declines were steeper where job tasks overlapped more heavily with AI capabilities, according to a Peking University index. Bloomberg also cited two independent Chinese studies that reached similar conclusions. It cautioned that the analysis cannot isolate AI’s causal effect because China’s broader economic slowdown may have disproportionately reduced office-job postings.
What problem it solves
Pi is designed for developers who need to adapt an agent to their own workflows rather than work within a fixed feature set. It provides interactive, automated, RPC, and SDK interfaces while leaving optional capabilities such as sub-agents and plan mode to extensions or packages.
Key features
Extensible agent harness — Customize the harness with extensions, skills, prompt te…
What problem it solves
Building LLM applications often requires connecting model providers, vector stores, tools, streaming, and agent capabilities. Rig provides a shared Rust interface for these components and includes a terminal coding agent that can be extended with plugins.
Key features
Unified model and vector-store interfaces — Use a shared interface for model providers and vector stores, wi…
What problem it solves
Managing multiple AI service sessions and translating them into compatible API interfaces can require repetitive account switching, custom routing, and protocol-specific integration. The project addresses these tasks with a desktop application that combines account administration, protocol conversion, request scheduling, and model routing.
Key features
Account dashboard and …
What problem it solves
Dodaios provides a locally installed, open-source way to make a personal AI assistant the control center for work on your computer. Its lead assistant coordinates models, coding agents, and computer use, while chats, memory, documents, and settings remain files on the computer that runs Dodaios.
Key features
Voice, chat, and phone interaction — The lead assistant supports wa…
What problem it solves
Managing subscriptions across multiple AI providers can require separate access paths and fragmented quota handling. This project provides a centralized gateway for distributing subscription quotas, supporting shared access and compatible use with native AI tools.
Key features
Subscription quota distribution — The gateway is designed to distribute and manage AI subscription …
What problem it solves
Developers need a unified way to integrate ChatGPT website capabilities into applications that expect OpenAI-compatible endpoints. They also need account-pool management, proxy configuration, import workflows, and a self-hosted interface for creating and reviewing media.
Key features
OpenAI-compatible endpoints — Provides compatible routes for image generation, image editing…
What problem it solves
Stock analysis often requires users to gather market quotes, news, technical indicators, financial information, and AI-assisted interpretation across multiple sources. The project addresses this by bringing supported-market data, analysis tools, AI agents, alerts, and review workflows into a locally focused desktop application.
Key features
Multi-market market data — The app…
What problem it solves
Large language models can be difficult for individuals to train, deploy, or understand because of their scale and the abstractions used by many frameworks. The project addresses this by providing native PyTorch implementations, compact model structures, open datasets, and staged workflows for pretraining, fine-tuning, preference optimization, tool use, and agentic training.
…
What problem it solves
AI agents may exceed their role, take unauthorized actions, or fail to follow organizational goals and policies. Teams need a way to test whether an agent behaves as intended across realistic and adversarial requests.
Key features
Business-aware simulations — Describe an agent, build a simulation environment for it, and exercise the agent with realistic requests that may exc…
What problem it solves
People often want an assistant that can participate in ongoing conversations, remember personal context, and use tools without requiring a cloud AI service. Jarvis addresses this with local-first voice and text interaction, personal memory, optional connected tools, and a setup process for locally controlled models.
Key features
Conversation-aware voice assistance — Jarvis m…
What problem it solves
Running and training models locally requires tools for model execution, fine-tuning, agent integration, and deployment across different model formats and hardware. The project brings these workflows into a desktop app, a web UI, and a code-based version.
Key features
Run and train models — Run and train language, diffusion, decision, embedding, audio, image, and video models…
What problem it solves
Personal and team knowledge is distributed across notes, conversations, trackers, and research, making it difficult to capture, organize, verify, and synthesize useful information. COG provides a markdown-based workflow in which agents can maintain a shared knowledge base, support recurring research and operational tasks, and verify selected work through independent checks.
…
What problem it solves
Configurations for the Codex desktop app, CLI, API providers, prompts, Sessions, Skills, MCP, and TOML files can become scattered across multiple locations and formats. Codex-X consolidates these recurring configuration tasks into a visual desktop interface so the current state is visible and common operations can be completed directly.
Key features
Visual prompt management …
What problem it solves
Serving large language models requires efficient memory management, flexible model execution, and practical integration with existing models and APIs. vLLM addresses these needs with inference and serving capabilities spanning batching, caching, quantization, distributed execution, and multiple hardware platforms.
Key features
Efficient memory and serving — vLLM manages atte…
What problem it solves
Generative AI systems and applications can contain security and safety weaknesses that require structured testing across models, endpoints, and complete applications. PyRIT addresses this with repeatable assessments, interactive exploration, and extensible workflows for red teaming these systems.
Key features
Repeatable command-line assessments — The scanner runs repeatable …
What problem it solves
Developers need Vietnamese speech synthesis that can run locally, produce natural reading styles, and support voice cloning without sending audio to a hosted service. They also need integration paths for SDK use, streaming applications, and an OpenAI-compatible API.
Key features
Instant voice cloning — Clone a voice from a short reference recording. The clipping pipeline aut…
What problem it solves
AI coding tools often expect particular provider protocols, making it difficult to use the models and providers a user prefers. A local gateway can translate between supported client APIs and provider APIs while preserving capabilities such as streaming, tool calls, reasoning, and images.
Key features
Provider protocol translation — The gateway translates Responses, Chat Com…
What problem it solves
AI agents often send large tool outputs, logs, files, RAG chunks, and conversation history to language models, increasing the input context that must be processed. Developers need to reduce that input while preserving important content, keeping compression local, and retaining access to original material when necessary.
Key features
Content-aware compression — A ContentRoute…
What problem it solves
The hidden system instructions that shape AI product behavior are difficult to inspect and change across models, products, and versions. This repository makes collected prompts available in a structured, product-oriented archive for study and comparison.
Key features
Broad product coverage — The archive covers prompts from major AI providers and products, including Anthropic…
What problem it solves
Image creation often requires switching between separate tools, manually translating visual ideas into parameters, and losing the context of an ongoing conversation. This project brings generation, editing, reference workflows, model comparison, and asset management into a DeepSeek Harness conversation and canvas.
Key features
Chat-Based Image Creation and Editing — Describe…