News · Leaderboard

Latest news, beside the leaderboard

Read what just changed, and see who is ahead. Those are the two doors on this page.

V
Vercel@vercel·Collected 45 min agoFresh · 6h
🛠️ Tooling

Vercel AI Gateway Adds Private Inference: Self-Hosted, Fine-Tuned and Prerelease Models via private/ IDs with the Same APIs, Auth and Logs as Catalog Models

Vercel announced on X and its changelog that AI Gateway now supports Private Inference: enterprises can register their own model endpoints as private providers and call them with private/<slug> model IDs through Chat Completions, Responses or Messages, with the gateway translating each request to the endpoint format. Private models are visible only to the team, credentials are stored encrypted, Zero Data Retention can be enabled per provider, the models work in coding agents such as Claude Code, Codex and Cursor, and requests can fall back to a catalog model if the endpoint fails.

Vercel AI Gateway Adds Private Inference: Self-Hosted, Fine-Tuned and Prerelease Models via private/ IDs with the Same APIs, Auth and Logs as Catalog Models
⚡ Key Takeaways
  • •Register fine-tuned, self-hosted or prerelease model endpoints as private providers and call them with private/<slug> IDs, so switching models only changes the model ID
  • •Supports the Chat Completions, Responses and Messages APIs, with the gateway translating each request to the endpoint format
  • •Private models are visible only to your Vercel team and not listed in the public catalog; credentials are encrypted and Zero Data Retention can be enabled per provider
  • •Works in coding agents connected to AI Gateway such as Claude Code, Codex and Cursor; logs, trace drains and reporting show private and catalog traffic together
  • •Optional fallback to a catalog model if the private endpoint fails; Vercel CLI 65.16.0 or later adds private-inference providers and models create commands
Read details→
S
Sarvam@SarvamAI·Collected 1h agoFresh · 6h
🛠️ Tooling

Sarvam Token Factory Opens to Everyone: India-Hosted Open-Weight Inference with One OpenAI-Compatible API for DeepSeek V4.1 Flash, Kimi K3.1 Pro and GLM 5.3

Indian AI company Sarvam announced on X and its blog that Sarvam Token Factory is now open to everyone. It serves Sarvam 105B plus GLM 5.3, GLM 5.3 Flash, DeepSeek V4.1 Flash 1017, DeepSeek V4 Pro, Kimi K3, Kimi K3.1 Pro and Gemma 4 31B from GPU clusters in India through one OpenAI-compatible API. Sarvam says it matches or beats leading global inference providers on speed and cost for most models, with DeepSeek V4.1 Flash launching at an introductory INR 14 per million input tokens and INR 55 per million output tokens.

Sarvam Token Factory Opens to Everyone: India-Hosted Open-Weight Inference with One OpenAI-Compatible API for DeepSeek V4.1 Flash, Kimi K3.1 Pro and GLM 5.3
⚡ Key Takeaways
  • •One OpenAI-compatible API serves Sarvam 105B, GLM 5.3, DeepSeek V4.1 Flash 1017, DeepSeek V4 Pro, Kimi K3.1 Pro, Gemma 4 31B and more from compute located in India
  • •DeepSeek V4.1 Flash has two endpoints: Interactive at about 200 tokens per second per user for coding agents (INR 25 input / INR 105 output per million tokens) and Standard at INR 14 input / INR 55 output
  • •Sarvam-measured figures: Sarvam 105B returns its first token in 0.2 seconds at 330 tokens per second for a single user; Gemma 4 31B returns its first token in 0.40 seconds on a 10K-token prompt, 5x faster than the only other provider serving it
  • •Kimi K3.1 Pro delivers up to 1.9x tokens per GPU versus InferenceX published results on coding tasks; DeepSeek V4.1 Flash serves 450B+ tokens a day with 95.7% of prompt tokens served from cache
  • •Sarvam runs India's largest B200 cluster and offers the same inference stack on-premise for enterprises, pitching data residency and no lock-in
Read details→
T
Tarun Sachdeva@tarunsachdeva·Collected 2h agoFresh · 6h
🛠️ Tooling

Traces Joins Vercel: AI Coding Agent Session-Sharing Platform with 16 Agent Adapters and Nearly 500,000 Shared Sessions Pauses New Orgs

Traces founder Tarun Sachdeva announced on X and the official blog that Traces is joining Vercel, where the whole team will keep working on how teams build software with agents. Launched in February as a CLI and platform for capturing and sharing AI coding agent sessions, Traces shipped Traces for Teams, adapters for 16 coding agents, Git integrations, Handoffs and Traces MCP in nine months, and thousands of teams have shared nearly 500,000 sessions. Nothing changes for existing users on day one, but new org creation is paused; deal terms were not disclosed.

Traces Joins Vercel: AI Coding Agent Session-Sharing Platform with 16 Agent Adapters and Nearly 500,000 Shared Sessions Pauses New Orgs
⚡ Key Takeaways
  • •Traces is a CLI and platform for capturing and sharing AI coding agent sessions, launched in February, and its whole team is joining Vercel
  • •In nine months it shipped Traces for Teams, adapters for 16 coding agents, Git integrations, Handoffs, Traces MCP and Surfaces
  • •Thousands of teams joined Traces and shared nearly 500,000 sessions
  • •Existing users can keep sharing sessions, using the CLI and adding users to existing orgs, but new org creation is paused; deal terms were not disclosed
  • •Vercel founder and CEO Guillermo Rauch wrote on X that traces are an essential new primitive of software development, capturing intent, prompts, reasoning, skills and tool use
Read details→
O
OpenAI@OpenAI·Collected 3h agoFresh · 6h
🚀 Release

OpenAI Upgrades ChatGPT for Healthcare: Computer-Use Agents for EHRs and Payer Portals, 20+ New Healthcare Plugins, GPT-6 Luna at 8.5 Cents per Task

OpenAI announced three updates to ChatGPT for Healthcare: computer-use agents that click, type and move across existing software such as EHRs and payer portals to finish multi-step work, improved trusted clinical search, and more than 20 new healthcare plugins. In OpenAI evaluations GPT-6 Astra earned 94.3% of rubric points on Stanford's HealthAdminBench and GPT-6 Luna 91.8% at 98% lower cost. CVS Health, Cigna, Abridge and Ambience are piloting or running it, and the capabilities are available today through the OpenAI API, ChatGPT for Healthcare and ChatGPT for Clinicians.

OpenAI Upgrades ChatGPT for Healthcare: Computer-Use Agents for EHRs and Payer Portals, 20+ New Healthcare Plugins, GPT-6 Luna at 8.5 Cents per Task
⚡ Key Takeaways
  • •Computer-use agents read screens, click and type to move information between EHRs, scheduling systems and payer portals, handling prior authorizations, referrals and discharge planning steps that APIs do not reach
  • •On EHR inbox tasks, 99% of 325 reviewed tasks were rated clinically safe, and physicians chose very or extremely helpful in 93% of 646 assessments
  • •On Stanford's independent HealthAdminBench, GPT-6 Astra earned 94.3% of rubric points and GPT-6 Luna 91.8% while costing 98% less at 8.5 cents per task (both at maximum reasoning effort, OpenAI-run evaluation)
  • •CVS Health cleared a months-long document-sending backlog over a weekend and is open sourcing an Agent Kit built on the Agents SDK; Cigna went from concept to deployment in under eight weeks
  • •More than 20 new healthcare plugins include UpToDate Expert AI, Metriport, Photon, Redox and Stedi; role-based access, audit logs and human approval steps are supported, HIPAA use requires a BAA, and data is not used for training
Read details→
ADSponsored
H
Hugging Face@huggingface·Collected 4h agoFresh · 6h
🛠️ Tooling

Hugging Face Partners With OpenRouter: Inference Providers Now Route Language Model Requests Through OpenRouter With the Same HF Token and Automatic Provider Failover

Hugging Face announced a partnership with OpenRouter: Inference Providers now routes language model requests through OpenRouter instead of its own router, picking a provider per request by live price, speed and reliability and retrying on the next provider if one fails, with no change to your code or HF token. Pricing is unchanged with no Hugging Face markup. Model pages gain Providers and Metrics tabs. Breaking change: you can no longer use your own provider API keys for language models.

Hugging Face Partners With OpenRouter: Inference Providers Now Route Language Model Requests Through OpenRouter With the Same HF Token and Automatic Provider Failover
⚡ Key Takeaways
  • •Language model requests now go through OpenRouter routing across 80+ providers, chosen by live price, speed and reliability, with automatic retry on the next provider inside the same call; image, video and audio generation stay on the existing path
  • •The router supports OpenAI Chat Completions, the Responses API and Anthropic Messages API, with setup guides for Claude Code, Codex, OpenCode, Pi and Hermes Agent
  • •Model id suffixes pick a policy: :fastest, :cheapest, or a specific provider such as :novita; an X-HF-Session-Id header keeps a conversation on the provider holding its prompt cache
  • •Pricing is unchanged with no Hugging Face markup; PRO users get $2 of monthly credits and Team/Enterprise organizations get $2 per seat
  • •Breaking change: bring-your-own provider API keys no longer work for language models (custom keys for image and video providers still do); linking an OpenRouter account to Hugging Face is the next step
Read details→
L
Liquid AI@liquidai·Collected 5h agoFresh · 6h
🚀 Release

Liquid AI Releases d1.2 and d1.2-ultra Omnimodal Decision Models With Audio Input and Five New Tasks; d1.2-ultra Tops the Artificial Analysis Decision Index v2.1 at 75

Liquid AI released d1.2 and d1.2-ultra, new versions of its decision models that take text, images and audio and add five new tasks: reading a value, extracting spans, locating objects in images, estimating counts and multi-select choices. d1.2-ultra scores 75 on the Artificial Analysis Decision Index v2.1, first place and two points ahead of the previous best decision model. Both are live on the Liquid API and bill input tokens only, at $0.02 per million for d1.2 and $0.042 for d1.2-ultra, the same price as the previous generation.

Liquid AI Releases d1.2 and d1.2-ultra Omnimodal Decision Models With Audio Input and Five New Tasks; d1.2-ultra Tops the Artificial Analysis Decision Index v2.1 at 75
⚡ Key Takeaways
  • •d1.2-ultra scores 75 on the Artificial Analysis Decision Index v2.1 (six benchmarks that also score answer probabilities), first among decision models; Liquid says no model scores higher than d1.2 for less money
  • •Five new tasks: read a text, number or date value (answering "not found" when absent), find every span of a kind, box one or all objects in an image, estimate a count with a range, and multi-select with per-option probabilities
  • •Audio input is new: you can ask about a recording (which language is spoken, whether a siren is in the background), and the question itself can be spoken
  • •Input-only billing with no output token charges: $0.02 per million input tokens for d1.2 and $0.042 for d1.2-ultra; images cost 1.5 tokens per 32x32 patch, about 1,536 tokens for 1024x1024
  • •Liquid-reported calibration error is 0.8 for d1.2-ultra and 1.2 for d1.2, versus 2.7 for Jev 1.14 and 3.7 for Perplexity Decider v1.3; new-task accuracy comes from Liquid internal test sets, not public benchmarks
Read details→
O
OpenAI@OpenAI·Collected 6h ago
📉 Price Cut

OpenAI Expands ChatGPT for Academic Researchers to Reach 200,000 Scientists, With 50,000 Free ChatGPT Plus Grants Opening November 23

OpenAI is expanding its ChatGPT for Academic Researchers program, aiming to give up to 200,000 researchers worldwide free ChatGPT access, double the goal set when the pilot launched in July. Applications for the next round open Monday, November 23, with 50,000 grants awarded first come, first served. Each grant covers the recipient and up to two collaborators with free ChatGPT Plus, access to frontier models and training. By default, data shared through the program is not used to train models.

OpenAI Expands ChatGPT for Academic Researchers to Reach 200,000 Scientists, With 50,000 Free ChatGPT Plus Grants Opening November 23
⚡ Key Takeaways
  • •The new round has 50,000 grants, each covering the recipient plus up to two collaborators, for up to 150,000 people; the overall goal of 200,000 researchers is double the July pilot target
  • •Selection moves from a lottery to first come, first served: the first 50,000 eligible applications are approved, and qualifying waitlisted researchers get activation emails before applications open
  • •The pilot offered ChatGPT Pro, but OpenAI found most participants stayed within Plus limits, so new grants are Plus to reach more teams
  • •Since July the pilot has enrolled 20,000 researchers in 33 countries; an OpenAI Economic Research survey of 1,800+ participants found they estimated the same work would take nearly twice as long without AI
  • •Program data is not used for training by default; OpenAI also gave $5 million to a new Simons Foundation math initiative, and biology labs can use the GPT-Rosalind academic plan
Read details→
O
OpenAI@OpenAI·Collected 7h ago
📊 Benchmark

OpenAI Launches First Embedded Assessments: Apollo Research, METR, Transluce and Goodfire Get Deep Access to Internal Models, Checkpoints and Training Data to Probe Alignment

OpenAI says it has signed agreements with four independent assessors, Apollo Research, METR, Transluce and Goodfire, for its first phase of embedded assessments. Depending on each evaluation, assessors can get unreleased models, training checkpoints, source code, training and evaluation data, monitoring systems and time with the responsible researchers, working on site at OpenAI to examine model alignment and whether safeguards work over an extended period. METR will look for misaligned behavior in OpenAI's largest reinforcement learning run to date. Assessors can publish independently, while OpenAI keeps the right to redact confidential information.

OpenAI Launches First Embedded Assessments: Apollo Research, METR, Transluce and Goodfire Get Deep Access to Internal Models, Checkpoints and Training Data to Probe Alignment
⚡ Key Takeaways
  • •Division of work: Apollo checks whether models could covertly work against developer intent and whether safeguards would catch it; Goodfire analyzes model internals with interpretability tools; METR assesses misaligned behavior in the largest RL run; Transluce reviews the quality and coverage of post-training alignment evaluations
  • •Access can include unreleased models, training checkpoints, source code, training and evaluation data, monitoring systems and researcher interviews; where the most sensitive IP such as weights or architecture is involved, OpenAI may pair assessors with its own researchers instead
  • •Assessors work on site at OpenAI on company devices; the Safety and Security Committee of the OpenAI Foundation board receives unredacted reports and resolves any remaining disputes
  • •Assessors publish independently, but OpenAI can redact confidential information and request responsible disclosure delays; Apollo and Goodfire are paid, while METR and Transluce take no fees but may accept API credits or compute
  • •OpenAI ties the program to its commitment under the White House Accord on Super Intelligence to independent assessments, and Transluce plans a follow-on assessment of about eight weeks in early 2027
Read details→
ADSponsored
V
Varun Mohan@_mohansolo·Collected 10h ago
🛠️ Tooling

Google Antigravity Doubles Context Window for Argon, Carbon and Hydrogen; Varun Mohan Says Compaction Drops 3.5x per User Turn

Varun Mohan of Google DeepMind (X bio: building @antigravity @GoogleDeepMind) announced on X that the context window for the Argon, Carbon and Hydrogen models in Antigravity has been doubled, and that context compaction per user turn should drop by 3.5x. The post drew more than 5,400 likes and was reposted by Sundar Pichai and Demis Hassabis. The exact window sizes were not disclosed.

⚡ Key Takeaways
  • •The context window for the Argon, Carbon and Hydrogen models in Antigravity has doubled, as announced on X by Varun Mohan of Google DeepMind himself
  • •The stated effect: context compaction per user turn drops by 3.5x
  • •The new token limits were not disclosed, and the post does not mention any pricing or quota change
  • •The post drew more than 5,400 likes and about 330,000 views, and was reposted by Sundar Pichai and Demis Hassabis
  • •Compaction is a common way long agentic coding sessions lose detail, so a bigger window lets the model keep more files and history in long sessions
Read details→
C
Chubby@kimmonismus·Collected 11h ago
📊 Benchmark

Remote Labor Index: Claude Fable 5.2 Tops Scale AI and CAIS Leaderboard at 41.25% Automation Rate, About Double Opus 5.5

The official Scale AI Remote Labor Index (RLI) leaderboard shows Anthropic's Claude Fable 5.2 in first place with a 41.25% automation rate, ahead of Opus 5.5 at 21.25%, GPT 6 Astra at 20.83% and Gemini Argon 4 at 17.08%. RLI, built by the Center for AI Safety and Scale AI, uses real paid remote freelance projects to test whether agents can deliver acceptable work end to end. The result drew wide discussion on X after @kimmonismus shared it.

⚡ Key Takeaways
  • •Official Scale AI RLI leaderboard: Fable 5.2 first at 41.25%, Opus 5.5 21.25%, GPT 6 Astra 20.83%, Fable 5.1 17.92%, Gemini Argon 4 17.08%
  • •The RLI score is the automation rate: the share of projects where the agent's deliverable meets a client-acceptable standard; the leaderboard notes there is no confidence interval for it
  • •Official scoring uses a private test set of 230 projects drawn from game development, product design, architecture, data analysis and video animation, together representing over 6,000 hours of real work valued at over $140,000
  • •The RLI site says the automation rate has doubled roughly every 3.3 months over the last 3 months; early agents on the board such as Manus 1.0 and gpt-5.2 sit around 2.5%
  • •At 41.25%, most projects still fail to meet the bar, so human review before delivery remains necessary
Read details→
T
Tencent Hy@TencentHunyuan·Collected 14h ago
🚀 Release

Tencent Launches Hy4 flash: Low-Cost Everyday Model at $0.070/$0.281 per Million Tokens, Close to Claude Haiku 5.5 in Official Blind Tests

On 2026-11-16 the official Tencent Hy account (@TencentHunyuan) launched Hy4 flash, calling it the cheapest model for everyday productivity. In Tencent's human blind tests, Hy4 flash scored 2.79 out of 4 for satisfaction, roughly level with Claude Haiku 5.5 (high) at 2.81, while costing 0.26 RMB per task versus 4.73 RMB, about 18x cheaper. DeepSeek-V4.1-Flash-1017 scored higher at 3.09 but cost 1.20 RMB per task. API pricing is $0.003 cached input, $0.070 input and $0.281 output per million tokens.

Tencent Launches Hy4 flash: Low-Cost Everyday Model at $0.070/$0.281 per Million Tokens, Close to Claude Haiku 5.5 in Official Blind Tests
⚡ Key Takeaways
  • •After the full Hy4 release on November 4, Tencent Hy adds the cheaper Hy4 flash, aimed at everyday office work and high-volume tasks
  • •API pricing: $0.003 cached input / $0.070 input / $0.281 output per million tokens
  • •Official blind test (4-point scale): Hy4 flash 2.79, Claude Haiku 5.5 2.81, DeepSeek-V4.1-Flash-1017 3.09, GLM-5.3-Flash 2.53; external models ran on their high tier
  • •Cost per task: Hy4 flash 0.26 RMB, Claude Haiku 5.5 4.73 RMB, DeepSeek-V4.1-Flash-1017 1.20 RMB; median end-to-end time 6.7 minutes, close to Haiku at 6.8
  • •Internal WorkIndex: Hy4 flash 42.1, above Hy4 preview at 39.7 and below Hy4 at 51.4, at about 24.5x lower cost per task than Hy4 (all Tencent internal figures)
Read details→
Q
Qwen@Alibaba_Qwen·Collected 21h ago
🚀 Release

Alibaba Officially Launches Qwen4-Max: New MoE Architecture, Native Vision, 1M Context, API Live on QwenCloud at $1.5/$4.5 per Million Tokens

On 2026-11-16 the official Qwen account announced the full launch of Qwen4-Max, following the Token Plan-only preview of November 11. Qwen says it is built on a new MoE architecture with native vision and a 1M-token context window, and that in internal evaluation it beats Qwen3.8-Max while using up to about 50% fewer tokens, with average cost per task down more than 60%. The API is live on QwenCloud at $1.5 per million input tokens, $4.5 per million output tokens and $0.075 per million for implicit caching, all below Qwen3.8-Max. It is also available in Qwen Studio, QwenWork and Qoder, and Qwen says open weights for smaller Qwen4 models will follow.

Alibaba Officially Launches Qwen4-Max: New MoE Architecture, Native Vision, 1M Context, API Live on QwenCloud at $1.5/$4.5 per Million Tokens
⚡ Key Takeaways
  • •Qwen4-Max moves from the November 11 Token Plan preview to a full launch, with the API now open on QwenCloud
  • •Per Qwen: new MoE architecture, native vision and 1M context; the QwenCloud model page lists 991K max input and 393K max output, with image and video input
  • •Efficiency: beats Qwen3.8-Max in internal evaluation with up to about 50% fewer tokens and more than 60% lower average cost per task (internal figures, no third-party benchmarks yet)
  • •Pricing: $1.5 input / $4.5 output / $0.075 implicit cache per million tokens, all below Qwen3.8-Max, with implicit caching 70% cheaper
  • •Compatible with the OpenAI and Anthropic APIs, with function calling, structured outputs, web search, batches and fine-tuning; Qwen says open weights for smaller Qwen4 models are coming
Read details→
O
OpenAI@OpenAI·Collected 22h ago
🚀 Release

OpenAI Launches a Science Blog: Physicist Nathaniel Craig and GPT-6 Astra Propose a Candidate Lattice Definition of the Standard Model, With Astra Leading the Detailed Proof

On 2026-11-15 OpenAI launched a new science blog. The first guest post is by Nathaniel Craig, head of the UCSB particle theory group and a visiting scientist at OpenAI. In an October preprint prepared with GPT-6 Astra, he proposed a candidate nonperturbative lattice definition of the Standard Model, addressing the phase (measure) problem for chiral fermions that had been open for more than 25 years. Craig says Astra turned his refinement idea into a working construction and took the lead on the detailed proof, with outside lattice experts reviewing it; he stresses it is a beginning that still needs expert scrutiny.

OpenAI Launches a Science Blog: Physicist Nathaniel Craig and GPT-6 Astra Propose a Candidate Lattice Definition of the Standard Model, With Astra Leading the Detailed Proof
⚡ Key Takeaways
  • •OpenAI opened a science blog; the first entry is a guest post by an outside physicist about using GPT-6 Astra in theoretical physics
  • •The problem: the Standard Model is chiral, and naive lattice versions produce unwanted doubler fermions; overlap fermions remove them, but the phase rule for the full Standard Model had been open for more than 25 years
  • •The method: represent the same gauge fields on a sequence of finer lattices, compare fermion states on neighboring grids and sum the phase corrections, giving the measure on the original lattice under a bound on field strengths at the grid scale
  • •Astra suggested how to carry the fields onto finer grids, showed the comparisons converge and led the detailed proof; lattice experts Maarten Golterman, Andreas Kronfeld and Yigal Shamir reviewed it and asked probing questions
  • •Caveat: Craig calls it an answer in principle rather than in practice, far from realistic lattice simulations, with properties such as reflection positivity still under study
Read details→
V
Vercel@vercel·Collected 1d ago
📉 Price Cut

Vercel Token Pass Now Includes $150 of Monthly Tokens for the Same $30 per Team (Up From $80), Covering Kimi K3, K3.1 Pro, GLM 5.3 and DeepSeek V4.1 Flash

On 2026-11-13 Vercel announced in its official changelog that Token Pass now includes $150 of tokens per month, up from $80, at the same $30 per team per month. The credit applies only to featured open-weight models on AI Gateway, including Kimi K3, Kimi K3.1 Pro, GLM 5.3, GLM 5.3 Flash and DeepSeek V4.1 Flash, and resets each billing period with no rollover. Token Pass remains an Early Access add-on for Pro teams with limited spots.

⚡ Key Takeaways
  • •Same price, more credit: $30 per team per month now includes $150 of monthly tokens, up from $80, about 5x the price
  • •Eligible models: featured open-weight models on AI Gateway, including Kimi K3, Kimi K3.1 Pro, GLM 5.3, GLM 5.3 Flash and DeepSeek V4.1 Flash; closed models are not included
  • •When credit runs out: requests using standard AI Gateway keys or OIDC tokens fall back to the team AI Gateway balance, while requests using Token Pass-scoped keys stop with no overage charges
  • •Credit resets each billing period with no rollover; Token Pass is an Early Access add-on for Pro teams, with a waitlist when spots fill
  • •Setup: run vercel token-pass in the Vercel CLI to connect a coding agent, or use a Token Pass API key in agents that support AI Gateway
Read details→
P
Patrick Collison@patrickc·Collected 1d ago
🚀 Release

X Debate: Stripe CEO Patrick Collison Asks How Humans Fare if AI Makes Labor Redundant, Since Smith and Montesquieu Thought Commerce Made Us Better; Post Drew About 1,270 Likes, 860 Bookmarks and 119K Views

Stripe co-founder and CEO Patrick Collison (@patrickc) posted on X at 09:04 SGT on November 15 that Smith and Montesquieu thought markets and commerce made us better, by creating an incentive to pursue self-improvement and to create things of value for others; in the positive AI visions where human labor is made redundant, he wonders how we fare. At the time of writing the post had about 1,265 likes, 121 reposts, 48 quotes, 180 replies, 864 bookmarks and roughly 119K views. Replies split: LSE professor Luis Garicano cited Viktor Frankl, arguing people are driven by a search for meaning that need not come from work; others argued status is the real driver, or worried that creative work loses meaning when anyone can generate it with AI. This is a discussion of opinions, not a fact or an announcement.

⚡ Key Takeaways
  • •The post: starting from Smith and Montesquieu's view that commerce rewards self-improvement and creating value for others, Collison asks how humans fare once AI makes labor redundant; about 1,265 likes, 864 bookmarks and roughly 119K views (post)
  • •Bookmarks are close to 70 percent of likes, suggesting many people saved it as a question worth returning to rather than a quick like
  • •LSE professor Luis Garicano replied with Viktor Frankl's "Man's Search for Meaning": purpose can come from raising kids or improving your neighborhood, and AI would not eliminate those larger purposes even if it eliminates work (reply)
  • •Other replies: some argued status is the real driver and commerce only a means; an engineer worried that when anyone can generate music and TV with AI, getting others to care about your work, and distribution, gets harder
  • •This is a discussion of opinions with no data or conclusion; Collison himself only posed the question and offered no answer
Read details→
A
Alon Amit@alonamit·Collected 1d ago
🚀 Release

X Debate: Alon Amit Asked Gemini 4 Carbon a Math Question on Low Thinking and It Dumped Pages of Scratchpad Into the Answer, Then Caught Itself: "I WAS WRITING DIRECTLY TO THE USER?!"; Post Drew About 3,300 Likes and 220K Views

Alon Amit (@alonamit), a VP at Intuit who posts about math, wrote on X at 07:46 SGT on November 15 that he asked Gemini 4 Carbon a math question and unintentionally left the Thinking level on Low; it started answering immediately, and after six screenfuls of math the screenshot he shared appeared. In it the model talks to itself: it had closed </thought> at the start and never reopened it, "I WAS WRITING DIRECTLY TO THE USER?!", blaming "effort level is 0.05 so my initial thought block closed immediately," noting it cannot delete already-emitted tokens, and telling itself "Okay, stop and breathe." Amit said it "went on like this for pages and pages." The post has about 3,291 likes, 99 reposts, 62 quotes, 149 replies and roughly 223K views, and many accounts reposted it as a joke. This is one user's experience; we found no public response from Google.

X Debate: Alon Amit Asked Gemini 4 Carbon a Math Question on Low Thinking and It Dumped Pages of Scratchpad Into the Answer, Then Caught Itself: "I WAS WRITING DIRECTLY TO THE USER?!"; Post Drew About 3,300 Likes and 220K Views
⚡ Key Takeaways
  • •The post: Alon Amit says he unintentionally left Thinking on Low, Carbon answered immediately, and after six screenfuls of math it started talking to itself and "went on like this for pages and pages," with about 3,291 likes, 862 bookmarks and roughly 223K views (post)
  • •The model's own explanation in the screenshot: it closed </thought> at the start and never reopened it, and "effort level is 0.05" made its initial thought block close immediately, so the scratchpad went straight into the user-facing answer
  • •It then noted it "cannot delete already-emitted tokens," told itself to "stop and breathe" and pivot to a clean answer, yet kept writing out its reasoning
  • •The debate: Low thinking saves time and cost, but this case suggests the line between thinking and answering can slip at low effort; many joked it is "the most human AI," while some said they would not put Low mode in front of customers
  • •This is one user's screenshot and opinion, not evidence of how Carbon usually behaves; we found no public response from Google, so treat it as discussion
Read details→
A
Anthropiccode.claude.com·Collected 1d ago
🛠️ Tooling

Claude Code Adds Experimental Code Mode: Claude Writes JavaScript to Batch Tool Calls in One Step, Opt-In From v2.1.323

Anthropic added Code Mode to Claude Code in v2.1.323 (published on GitHub on November 12) and shipped an official docs page for it. With it on, Claude stops requesting tools such as Bash, Read and Edit one at a time; instead it writes a piece of JavaScript that runs commands in parallel, feeds one result into the next call, and returns only the output the model needs. The feature is off by default and turned on with the environment variable CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1. It works only in an interactive claude session in a terminal, on the native build, with models such as Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.2. v2.1.325 on November 13 fixed a batch of Code Mode bugs.

⚡ Key Takeaways
  • •How to turn it on: set CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1 in your shell or ~/.claude/settings.json and confirm on first launch; the variable is ignored in project settings, so a cloned repo can't turn it on for you (official docs)
  • •How it works: Claude writes JavaScript that calls its tools; one piece of code can run commands in parallel and chain results, while the UI still shows the tool calls and Ctrl+O reveals the code
  • •Limits: interactive terminal sessions only; not -p, the Agent SDK, the VS Code panel, the Desktop app or cloud sessions, and not through a gateway, Bedrock, Google Cloud's Agent Platform or Foundry; unavailable to HIPAA and zero-data-retention organizations
  • •Permissions: every tool call still goes through deny/ask rules and PreToolUse hooks; in auto mode the classifier reviews the whole code, not each call; foreground code stops after 10 minutes, background code can run up to 2 hours
  • •Admins can set the variable to 0 in managed settings to block it org-wide; v2.1.325 fixed Code Mode bugs around worktrees, lost input at permission dialogs and resumed sessions starting work by themselves (changelog)
Read details→
C
Cloudflareblog.cloudflare.com·Collected 1d ago
🛠️ Tooling

Cloudflare Open-Sources Kitesurf, Its Browser Engine for AI Agents That Runs on Workers Without a Browser Container, Under Apache 2.0 With CDP/Puppeteer Support

Cloudflare announced on its blog and developer changelog that Kitesurf is now open source on GitHub under the Apache 2.0 license. Introduced in August as a stateless browser for AI agents, Kitesurf runs entirely on Cloudflare Workers: it parses HTML, applies CSS, executes page JavaScript and renders screenshots without Chromium or a browser container, and it speaks the Chrome DevTools Protocol so Puppeteer can connect. Developers can now run it locally, change it, and deploy it to their own Cloudflare account instead of the managed Browser Run APIs; self-hosting requires a Workers Paid plan, and the engine has no built-in authentication.

Cloudflare Open-Sources Kitesurf, Its Browser Engine for AI Agents That Runs on Workers Without a Browser Container, Under Apache 2.0 With CDP/Puppeteer Support
⚡ Key Takeaways
  • •Kitesurf is open source on GitHub (cloudflare/kitesurf) under Apache 2.0; you can read the code, run it locally, modify it and deploy it to your own account (Cloudflare Blog)
  • •Positioned as a lightweight browser for AI agents: runs entirely on Workers with no browser container, aiming to use less CPU and memory than Chromium
  • •Speaks the Chrome DevTools Protocol, so Puppeteer and other CDP clients connect directly; Cloudflare also suggests pointing an AI harness at the local CDP port
  • •Stack: Blitz (HTML/DOM/layout/painting), Stylo (CSS), Boa (JavaScript fallback inside WebAssembly), Linebender (text and 2D graphics), Krilla (PDF)
  • •Self-hosting needs a Workers Paid plan (it uses Dynamic Workers), and the engine has no built-in auth, so Cloudflare says to put Cloudflare Access or another auth layer in front; the project is still experimental
Read details→
G
Google Clouddocs.cloud.google.com·Collected 1d ago
📉 Price Cut

Google Cloud Tightens the $300 Free Trial Credit: From November 16, Accounts That Sign Up or Upgrade to Paid Can No Longer Spend It on Gemini Platform (Formerly Vertex AI)

Google Cloud's official "Free Google Cloud features and trial offer" documentation now states that the $300 Welcome credit can't pay for Gemini platform (formerly Gemini Enterprise Agent Platform / Vertex AI) services if you sign up for a free trial or upgrade to a paid account on or after November 16, 2026. The same page says the credit already can't pay for Gemini API costs in AI Studio, or for generative AI partner models offered as managed APIs. For new developers hoping to call Gemini models on trial credit, the remaining routes from tomorrow are the Gemini API free tier or a Prepay billing plan.

⚡ Key Takeaways
  • •Effective date: if you sign up for the free trial or upgrade to a paid account on or after November 16, 2026, the $300 credit can no longer pay for Gemini platform (formerly Vertex AI) services (Google Cloud docs)
  • •Gemini API costs in AI Studio were already excluded from the credit; Google points users to the Gemini API Free tier or a Prepay billing plan
  • •Generative AI partner models offered as managed APIs (model as a service) can't use the credit either
  • •Everything else stays: $300 credit valid for 90 days; after upgrading to paid, remaining credit is kept but must be used within 90 days of signup
  • •The wording names upgrading to paid as a trigger too, so existing trial accounts planning to spend credit on Gemini should watch when they upgrade
Read details→
P
Pope Leo XIV@Pontifex·Collected 1d ago
🚀 Release

X Debate: Pope Leo XIV Calls Transhumanism and Posthumanism a "Mirage" That Must Not Become "a Privilege Reserved for Millionaires Seeking Immortality"; His Post Drew 61K Likes and About 3.8M Views as roon and Others in the AI Crowd Weigh In

Pope Leo XIV (@Pontifex) spoke on November 13 at the Santo Toribio de Mogrovejo Catholic University in Chiclayo, Peru, and at 10:10 SGT on November 14 posted a passage from it on X: the promises of transhumanism and posthumanism are multiplying, ideological currents holding that technology will sooner or later eliminate disease, pain, aging and even death; scientific progress "must not be confused with a privilege reserved for millionaires seeking immortality," and "a limitation is not a manufacturing defect, but rather the very condition of human grandeur." The post has about 3.8M views, 61,476 likes and 7,847 reposts. At 06:31 SGT on November 15, X user roon (@tszzl) quoted it, calling it "pretty amazing" that an office at least 2,000 years old is "around to opine on the advent of posthumanism," drawing about 1,065 likes. This piece covers opinions and discussion; the speech does not name any AI company or product.

⚡ Key Takeaways
  • •The Pope's post: transhumanism and posthumanism hold that technology will eliminate disease, pain, aging and even death; scientific progress "must not be confused with a privilege reserved for millionaires seeking immortality," with about 61,476 likes, 7,847 reposts, 1,507 replies and roughly 3.8M views (post)
  • •The full speech (Holy See Press Office) also says "technology is not always neutral" and can bear "the imprint, interests and ambitions of those who design and control it"; it calls submission to the technocratic paradigm that reduces reality to data and treats efficiency as the supreme value "the most deadly risk," and urges seeking "that wisdom which neither machines nor computer programs can possess"
  • •X user roon (@tszzl) quoted it: "it is pretty amazing that the Pontifex Maximus of Rome, an office that has existed for 2000 years at least, is around to opine on the advent of posthumanism," with about 1,065 likes and 54 replies (quote post)
  • •The point of the debate: visions common in AI circles, such as posthumanism, radical life extension and AGI reshaping the human condition, are now being answered head-on at the top of religious and ethical institutions
  • •These are opinions: the speech names no AI company, model or product and sets no concrete policy; treat it as discussion
Read details→