Mastra announced Mastra Gateway, a hosted gateway on its platform that gives users one API key and bill for accessing models from providers including Anthropic, OpenAI and Typesafe AI. Mastra says the service offers more than 500 language, voice, video and decision models, along with automatic failover between inference providers, regional routing between the US and EU, per-key model and spending limits, consolidated billing and observability for inference calls. The company said the gateway is built on the model router its Mastra agents have used for two years. These model-count and feature claims come from Mastra and have not been independently verified.
The Ramp AI Index found that AI token prices fell 21.2% over six weeks while token volume rose 1.5% to a record high. Spending fell 6.3% from its October peak, although it still rose 12% month over month; OpenRouter data showed a similar pattern of sharply higher usage and slower spending growth. Stripe data also indicated that revenue at tracked AI companies remained high but was decelerating.
Altimeter investor Pauline Yang said the firm added to its Anthropic position in the offering because improving compute economics transformed its investment thesis. Speaking with Bloomberg Technology ahead of Anthropic’s shares listing on Nasdaq, Yang also said she sees room for both Anthropic and OpenAI to become multi-trillion-dollar companies, an investor assessment rather than a reported fact.
a16z announced the second cohort of its FDE Fellowship, an eight-week program for Forward Deployed Engineers and Applied AI leaders starting mid-January 2027. Dinners and events will be held every other week, mostly in San Francisco and New York, while fellows may be based anywhere. Applications are open through December 31, 2026, and will be reviewed on a rolling basis for a limited number of spots.
Rest of World reports that Scale AI's Outlier advertises up to $180 an hour for U.S. physicians designing cardiac scenarios and reviewing AI-generated reports, while similar work in the Philippines and Tanzania is advertised at much lower rates. Interviews with more than 30 workers describe a recurring pattern of rising demands, shorter target times, disappearing tasks and falling pay as projects progress. The report also says a proposed settlement involving Scale AI workers received preliminary approval in San Francisco, although the company denies wrongdoing.
AMD shares fell more than 7% after Hunterbrook Capital disclosed a short position and released a report alleging circular deals, excessive customer commitments and business ties to Chinese companies on a U.S. blacklist. Hunterbrook estimated $28 billion in credit support for partners and $48.8 billion in supplier commitments; it also alleged that AMD issued stock warrants to OpenAI and Meta. AMD disputed the report's characterization of its commercial practices and partnerships and said it complies with U.S. export-control law.
A report states that ChatGPT has launched a live dictation feature and removed a 10-minute duration limit. The supplied report consists only of a headline and does not provide further details or a new maximum duration.
Arena introduced its Coding Index to evaluate coding agents using real coding work. Arena reported that GPT-6.2 Astra (Max) scored 1,338 points and ranked third, while its overall Agent Arena Index score was 1,291 points at sixth place. On the Coding Index, Arena said GPT-6.2 Astra led GPT-6 Astra (Max) by 50 points with a 62% lower median cost per task.
Anthropic priced its IPO offering at $83 per share, above the preliminary pricing range, valuing the maker of Claude at about $1.5 trillion. The report said Anthropic reached the public markets before OpenAI. Employees may sell up to 10% of their shares on the first trading day but cannot sell more until the company reports full-year 2026 earnings early the following year. The report also said the stock was likely to trade significantly higher when it opened, a forecast rather than a reported result. It raised the possibility that newly wealthy employees could leave, but no employee exodus was confirmed.
Nvidia spent $58.1bn buying shares in other companies in the nine months ending 25 October, including stakes in OpenAI and Anthropic, according to its quarterly results cited by Fortune and The Next Web. The company also reported $32bn in future commitments to neocloud providers and $118bn in credit guarantees, including $105bn for the PORTS-Pike data centre in Ohio, which OpenAI plans to lease for 20 years. Nvidia said some guarantees take effect only after a customer has secured financing. The stakes generated $1.8bn in net gains during the quarter.
The Verge reports that Anthropic’s IPO raised $81.3 billion and produced a valuation above SpaceX’s, making it the largest IPO on record at the time. The company’s Public Benefit Corporation structure, founders’ controlling voting rights and Long-Term Benefit Trust shield it from ordinary shareholder pressure while concentrating substantial decision-making authority. Separately, 36氪, citing Sina Finance, reported that Abu Dhabi’s MGX gained $20 billion from Anthropic’s record offering.
Kixelated argues that WebRTC’s low-latency design can discard late speech packets, causing voice AI systems to lose or mishear user prompts and ask users to repeat themselves. The author proposes Media over QUIC as a better transport and demonstrates Pipecat support for MoQ, while noting that packet loss elsewhere in the pipeline remains outside the demo’s control. OpenAI WebRTC experts Sean DuBois and Justin Uberti reportedly contacted the author after his earlier criticism.
One report credited Codex with helping secure Ruby, with assistance from the team behind the statement. The cited researcher separately said they independently found bugs enabling a universal pre-authentication remote-code-execution attack on Ruby on Rails. Another report said Ruby disclosed seven vulnerabilities in its regular-expression engine and string case-conversion functionality, including flaws for which proof-of-concept code could trigger remote code execution. Fixes were released in Ruby 4.0.10, 3.4.13 and 3.3.14.
In a first-party retrospective, the author describes ThinRead 0.2.2, an MIT-licensed macOS reading-and-writing application built with Electron, React and TypeScript. The tool stores books and notes locally, verifies generated quotations against source text, validates streamed AI output before writing, and uses local BGE-small-zh embeddings for cross-book connections, while supporting multiple cloud and local model providers. The author also reports an import pipeline covering eight file formats, webpages, OPDS, WeChat Reading, Readwise and Kindle highlights, plus macOS OCR for scanned PDFs.
AdClarity estimated that US advertisers spent $70.7 million on ChatGPT ads in October, up 127% from $31.1 million in April. Estimated spending across 24 countries reached $110.6 million, with the US accounting for nearly 64%; programming and developer software was the largest category at $27.9 million.
A commentary says Anthropic researcher Jacob Coxon resigned after warning that Anthropic and OpenAI were racing toward self-improving superintelligence and “gambling with our lives.” Anthropic’s head of alignment science, Evan Hubinger, said Coxon was correct that AI could kill all humans, while CEO Dario Amodei told CNN that he agreed with Coxon more than he disagreed with him. The commentary argues that recursive self-improvement remains speculative and that pursuing it could produce systems that are harder for humans to understand and control, even without an intelligence explosion.
O’Reilly Radar argues that AI is moving from being built by humans to participating in its own development, a pattern researchers call recursive self-improvement. It cites Anthropic’s vendor claim that Claude writes more than 80% of merged code, Google’s AlphaEvolve finding a way to speed up Gemini training, and OpenAI’s automated research intern carrying out research tasks under human direction. The article frames these deployments as practical examples rather than proof of autonomous, open-ended self-improvement.
AIX Finance, republished by 36Kr and TMTPost, reports that AI video generation is moving toward lower per-second pricing and increasingly interchangeable model performance. Google made Gemini Omni 1.2 Flash generally available with 15-second generation and multi-turn conversational editing, while pricing for 720p output reportedly fell to $0.07 per second; Shengshu Technology also introduced Vidu Q4 Preview at a promotional rate of RMB 0.09 per second. The report examines whether large platform ecosystems can absorb continued inference costs even as OpenAI reportedly shut down Sora and no video-model company has publicly disclosed profitability.
Ally Bank, JPMorgan Chase and Citi warned customers that actions taken by AI agents with authorized financial-account access may be treated as the user's own acts. Ally Bank said users bear responsibility when AI services make mistakes, hallucinate or misunderstand instructions, while JPMorgan Chase said authorized transactions may be impossible to reverse if customer information is abused. Citi added website and app banners directing users to AI-assistant risk information.
A 36Kr report and commentary by Kuaidao Qingyi connect sharply rising AI token demand with difficulty for model companies to generate profits. OpenRouter data cited by 36Kr show tokens routed through the platform rising about 250-fold to 128.4 trillion over 20 months. In the second week of September, open-weight models handled 72% of those tokens but accounted for only 27% of platform spending. The commentary reports that Salesforce introduced role-based token budgets, Apple reduced some engineering quotas, and Uber imposed per-person, per-tool limits after exhausting its annual budget in four months. It also argues that outcome-based pricing remains difficult for general-purpose AI because business results can be hard to define and attribute among AI, people and other factors.
Anthropic began trading on Nasdaq on November 19 at $83 per share, raising about $81.3 billion through its public offering while a concurrent $26 billion private placement brought total financing to roughly $107.3 billion, according to TMTPost and 钛媒体. The private placement drew major chip and cloud suppliers, while filings cited roughly $900.9 billion in computing and infrastructure commitments, highlighting how investors, customers, lenders and vendors are becoming tightly connected. 虎嗅 AI and 36氪 AI also examined Amazon’s valuation gains, employee wealth estimates and the financing risks created by long-duration commitments.
Sellbird 0.4.0 is the new name for the open-source outreach platform previously called Harvey, with existing configurations, data, and dashboard sign-in carried over. According to the project README, it runs locally through Claude Code CLI or Codex, and can find prospects, verify email addresses, draft personalized outreach, send approved emails, and handle replies. A hosted version with sending-domain, inbox, and warmup setup is available in early access.
A Harvard Business School study led by Mark Egan analyzed 114 billion Fiserv transaction records from 2021 through April 2026, covering about 174 million consumers and paid subscriptions from OpenAI, Anthropic, Google, and xAI. TechNews reported that about 2.9% of U.S. consumers paid for major generative AI services as of April 2026, with paid penetration of about 2.4% for OpenAI and 0.5% for Anthropic. The study also found high price sensitivity, one-year retention of about 45% for OpenAI and 30% for Anthropic, and that the share of Anthropic subscribers who also subscribed to OpenAI fell from about 45% in June 2024 to below 30% by April 2026.
A V2EX user scanned model-level listings at 545 relay providers, including 509 that offered GPT models, and found that 31 listed any GPT-6.2 variant while 478 did not; 413 of the latter remained on GPT-6.1. The author cautioned that the survey reflects a limited snapshot of public pricing endpoints, not provider-wide verification, and said relay services may need time to connect upstream access and configure testing.
OpenAI says its cloud-based Performance Factory uses Codex agents to investigate ChatGPT performance regressions, propose fixes for engineer review, and check after deployment whether performance returned to baseline. A separate review step tries to disprove proposed diagnoses. OpenAI reports that, in just over three months, its Regression Workflow tracked more than 1,700 investigations and identified more than 600 root causes; in a manual review of 100 root-cause suggestions, 70% were graded as accurate. A secondary Chinese report said about 50 production fixes adopted or referenced an agent’s proposal, while only around 15 pull requests written by Codex were merged with little or no human modification.
A Juejin tutorial presents Vercel AI SDK 6 as a unified abstraction for switching between AI providers while keeping surrounding application code unchanged. Its first installment demonstrates project setup, text generation with generateText, streaming output with streamText, and basic use of OpenAI and Anthropic provider packages; the author's productivity claims remain anecdotal.
A reviewer tested a model identified as GPT-6.2 Astra on image generation, 3D modeling, SVG and controller-interface tasks, and a large game project. The reviewer found focused results generally strong, including capable controller handling, an interactive experience, and a reference-based SVG that was praised for both appearance and function. Completion times were also described as fast. By contrast, the loosely specified game project had performance, functionality, playability, and visual-realism problems. The reviewer said its game was less playable and complete than results produced by Fable 5.2 and Opus 5.5, while describing its overall autonomy and engineering delivery as well behind Claude.
Securities Times supply-chain reports cited by Android Authority and IT之家 AI say Foxconn subsidiary FIH Mobile has won an order to assemble OpenAI’s first smartphone. The device is reportedly expected to use a custom MediaTek chip and an AI-first experience, while Android Authority says it could launch in late 2027 or 2028. Neither OpenAI nor FIH has officially confirmed the order, and the operating system, order scale and production schedule remain undisclosed.
Former OpenAI and Anthropic researcher Jacob Coxon said he plans to sell his OpenAI equity and donate all proceeds to AI safety causes. He called for the race toward superintelligence to be paused or drastically slowed, and said he intends to remain independent while using his platform to raise awareness of AI policy. TechNews reported that the size of his stake, the sale schedule and the recipient organizations were not disclosed.
A Juejin analysis reports that TypeSafe AI released Jev on September 15, 2026. Jev accepts state and typed questions and returns option probabilities for classification tasks without generating free-form text. Version 1.14 added the Box primitive for image localization, while JevBench Capability fell from 77.1 for version 1.13 to 75.0; Calibration remained near 90, declining from 90.6 to 90.1. The analysis says Jev offers low latency but is mid-to-high priced among decision models, with some faster or more accurate competitors. It also notes that type safety confines answers to supplied options but does not guarantee they are correct.
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…