OpenRouter BlogSelf-hosted LLM is only cheaper than API when GPU load is high enough. Taking the price on October 8, 2026 as an example, a 27B model needs 6-12% load on rented RTX 3090 to balance with API, while DeepSeek V4.1 Flash never balances on four H200s.
Codex releases00GitHub Copilot has released version 0.170.0-alpha.2, offering features such as code review, vulnerability detection, and key protection.
OpenRouter BlogRunning the same tasks on OpenRouter, GPT-6.1 Sol passed 180/180 tests in 60 synthetic tasks, DeepSeek had the lowest cost passing 157/180 times, but the correlation with the public ranking is low (Spearman correlation coefficient 0.109). Need to check the number of task repetitions, the number of answers, and billing verification.
OpenRouter BlogThe OpenRouter Agent SDK, by enabling doom-loop detection, prevents infinite loops in six tested models when the third identical tool call is received; combining stepCountIs(40) and maxCost(x) can limit the number of runs and cost; the credit limit of the API key can control total expenditure, these measures effectively reduce the number of tool calls (from 314 to 64) and costs (from $0.29 to $0.05).
Nvidia CEO Huang Renxun has returned to work at Nvidia and apologized to employees over the recent controversy; Anthropic's IPO demand exceeds several times, planning to raise up to $90.4 billion through IPO and private financing; Google launches Gemini Decisions API, charging $0.10 per million input tokens.
Chip startup Fractile raises $720 million in funding, valuing it at $7 billion post-investment, becoming one of the highest-valued startups in the UK. Its self-developed AI chip with custom memory, Anthropic intends to be among the first customers.
Alan Waxman, CEO of Sixth Street, expresses concerns about credit expansion and speculative behavior in the AI field, believing the economic efficiency of AI units is difficult to assess, and lending institutions have reduced financing for AI infrastructure projects.
Santosh Janardhan, Meta's Infrastructure Director and Joint Engineering Director, is expected to leave the company in early 2027. Meta is investing tens of billions of dollars in AI and data center construction, and this departure constitutes a major change in the company's management.
How the accounts differ:TechCrunch reports the departure is planned for next year, while 36Kr states it's in early 2027.
AWS Machine Learning BlogAmazon Bedrock AgentCore has launched the Web Search API today, allowing developers to search the web index operated by Amazon with a single API call. The results include text summaries, page titles, URLs, and publication dates.
AWS Machine Learning Blog5858Amazon SageMaker HyperPod detects silent data corruption (SDC) by adding three checks in the PyTorch DDP training loop, including non-finite gradient checks, loop replay audits, and weight fingerprint comparisons. When issues are detected, suspicious nodes are isolated and replaced by HyperPod, having been tested and verified on a 4 ml.p5.48xlarge instance cluster.
AWS Machine Learning BlogDevelopers use OpenAI Codex for code exploration and engineering work through Amazon Bedrock. Amazon Bedrock supports Codex inference and can be integrated with AWS Identity, IAM, and observability practices. It is recommended to start with Native AWS Access for small-scale pilots, use GPT-5.6 Sol for validation, authenticate through IAM Identity Center, and configure model permissions and observability targets.
88% of surveyed developers support the creation of AI usage policies for FreeBSD, with 91% agreeing that humans should bear responsibility; an AI working group has been established, with a report expected to be submitted in the fourth quarter; simultaneously, 51 progress items in areas such as security compliance, virtualization, and desktop experience are being advanced.
Databricks BlogAlasdair Anderson, a Solutions Architect at Databricks, shared his daily work routine, covering technical construction and team coaching; he emphasized how platforms like Lakebase and Agentbricks help financial industry clients achieve second-level data processing and business process reshaping; 'Customer First' and 'Pragmatic Truth' in the company culture are reflected in practices such as analyzing industry reports and real-time risk modeling, while open and transparent internal collaboration also promotes global teams in efficiently solving complex challenges.
Databricks BlogDatabricks detects credential-driven attacks by analyzing audit logs and Genie, finding stolen credentials listed from new IPs in secret scopes, directories, and repositories within minutes, then reading secrets containing service principal credentials, and subsequently having the service principal create functions and join admin groups, these non-linear event sequences reveal intrusion behavior.
Google DeepMind has launched the AlphaGenome 2 model, which can predict the effects of DNA variations on specific cell types, nearly doubling its ability to identify gene expression variations to 30% (from 16%), and accurately predicting the direction of gene expression changes to 43% (from 23%). It supports cross-modal data inference and has opened API access to non-commercial research.
Google launches the Gemini Decisions API, driven by the Gemini Decision1 model, allowing developers to evaluate up to 1000 questions for multimodal input, scoring preset options in a single forward pass and returning probabilities and confidence levels, with end-to-end latency of 150–250 milliseconds, and a fee of 0.10 USD per million input tokens.
Anthropic's IPO demand has reached several times the number of shares available for issuance, primarily driven by commitments from large anchor investors. The company and its co-founders plan to raise a maximum of $90.4 billion through IPO and private placements, which, if completed, will surpass the SpaceX IPO and become the largest IPO in history. Anthropic's fully diluted valuation target is approximately $1.4 trillion, lower than some investors' previous expectations.
Supermicro Computer announces the start of shipments for the H15 server series, equipped with the AMD 6th Generation EPYC 9006 series processors (code name Venice), providing up to a 33% core count increase, 2 times PCIe bandwidth and 1.6TB/s memory bandwidth, setting multiple world records in the SPEC CPU benchmark test.
Nothing CEO Pei Yu showcased the appearance of the company's first AI device, equipped with LED indicators and physical buttons, and is expected to adopt a screenless design. The device resembles the science fiction work 'Lost in Space,' and the buttons may be an entry point for interacting with AI. Pei Yu has previously stated that the company will focus on AI, and Nothing raised $200 million in funding for AI development in 2025, with the first AI product expected to be released in December 2026.
Hugging Face Daily Papers6565MACaQ is a post-training quantization method for MXFP4 W4A4, which reduces costs while maintaining accuracy in language and diffusion models by using online activation transformations and Hessian-aware refinement to independently quantize each linear layer. MACaQ achieves only a 2.3% end-to-end overhead increase on Qwen3-32B with vLLM, has a lower perplexity than DuQuant++ and MR-GPTQ, and reaches the best perplexity of Llama-3-8B in 11 times shorter quantization time.
Browser UseBrowser Use Cloud supports self-managed residential proxies and Bring Your Own (BYO) proxies. Users can provide connection details of third-party providers through BYO proxies when creating proxy runs or cloud browsers to control the provider, exit location, and session settings. Users need Python 3.10 or higher, access to third-party proxy providers, and a Browser Use API key.
AWS Machine Learning Blog3838Google is gradually rolling out the CarPlay integration feature to iOS users worldwide, covering all languages supported by Gemini Live. Users with devices like iPhone 18 Pro can use Gemini's voice mode for hands-free conversations on the CarPlay in-car screen, with chat history synced to the iPhone app. Users can connect to apps like Gmail and Google Calendar to get daily briefs or ask for destination information, practice speeches, etc., with conversations automatically synced to the phone.
The Verge · AI6565United Airlines has stated it will not allow personal AI agents to book tickets for privacy and regulatory issues, although no large number of users have yet attempted to use AI agents for bookings. The company's decision will limit the potential applications of new personal AI agents like Meta Muse and OpenAI Dots. CEO Scott Kirby said he does not want to create new middlemen. Other industry companies like eBay and several banks have also taken similar restrictions, and United Airlines said it may reconsider its decision based on future advancements in AI agents.
Modal Blog6060
Simon WillisonSame story · 7 outletsGoogle has released the multimodal decision-making model Gemini Decisions API, supporting text, image, audio, video, and PDF, similar to Jev and OpenAI's decision models but more confident; the price is 10 cents per million input tokens, output is free, more expensive than Jev (4.2 cents per million tokens) and OpenAI (same price)
MarkTechPostFeaturedSame story · 7 outlets7676AWS Machine Learning Blog1212AWS re:Invent 2026 AI/ML themed symposium will be held from November 30 to December 4, offering over 140 workshops covering full-stack content such as AI chips, model training and inference, and agent development management.
Simon WillisonSame story · 2 outlets5858AWS has released the aws-emulator emulator, which allows users to clone various AWS services locally for testing without using paid API keys. The emulator is written in Rust, and the project received 29,684 commits in the first three weeks. Currently, it requires Docker to run, but developers hope to provide a cross-platform single binary version in the future.
How the accounts differ:Simon Willison states the simulator uses Rust and received 29,684 submissions in the first three weeks; MarkTechPost mentions it is built by an AI agent and supports over 200 AWS services.
The DecoderFeaturedSame story · 7 outlets7878Google DeepMind launches Gemini Decisions API, targeting yes/no, multiple choice, and scale scoring for closed-form judgments. The underlying model, Gemini Decision 1, can handle up to 1000 independent questions at once, covering text, images, audio, video, and PDF.
MarkTechPostSame story · 5 outlets6060Google DeepMind releases AlphaGenome 2, a DNA model that predicts the effects of specific cell type mutations. The model reads 1 Mb DNA and target cell state, significantly improving single-cell eQTL prediction performance through conditional sequence to functional architecture, increasing recall from 23% to 43% on the TenK10K benchmark, and can adapt to unseen datasets without fine-tuning.
AWS Machine Learning Blog3838DXC Technology has built a multi-model AI infrastructure on Amazon Bedrock, accessing Anthropic Claude, Amazon Nova, Meta Llama, and open-source model families through a unified API, and using Amazon Bedrock Guardrails to implement content filtering and PII anonymization.
AWS Machine Learning BlogAWS CloudTrail captures the UpdateAgentRuntime API call event for AWS Agent Registry in Amazon Bedrock AgentCore, Amazon EventBridge routes the matched event to an AWS Lambda function, and the Lambda function queries the current tools from the MCP server and updates the matched Registry record.
TechCrunch · AICoreWeave adjusted the way it charges for AI prosperity, requiring customers to pay in advance, pledge specific contracts for loans, and seek short-term transactions to charge higher prices. The company reported $3.66 billion in revenue for the third quarter, a year-on-year increase of 168%, but a net loss of $709 million, with quarterly interest expenses of $846 million.
TechCrunch · AI6565The boom in data center construction is facing real-world obstacles, including local opposition and approval delays, as well as power struggles. Billions of dollars in AI infrastructure investment are at risk, and it is becoming important to distinguish between actual and speculative projects. Brent Nelson, Senior Director of Market Strategy at Ascend Analytics, believes that most proposed projects may not be built, and if the AI infrastructure bubble bursts, developers of wrong tenants will be in trouble.
TechCrunch · AISame story · 2 outlets3535Meta informed TechCrunch that Santosh Janardhan, the company's long-term infrastructure head, will be leaving, to be replaced by Dave Brown, who joined Meta as an infrastructure vice president this year. Janardhan had originally planned to leave next year and will take on an advisory role at Meta Compute. Meta did not specify the reason for Janardhan's departure.
How the accounts differ:TechCrunch reports the departure is planned for next year, while 36Kr states it's in early 2027.
