☀ TL;DR

Anthropic releases Opus 5.5 with lower output pricing

Anthropic released Opus 5.5, cutting output pricing to $20 per million tokens from $25 and claiming state-of-the-art coding and knowledge work results.

OpenAI said its model solved Navier-Stokes, but mathematicians found the proof covered only an unrealistic variant.

A paper reports open-weight inference costs about one-fifth of closed models, with A100s remaining economical.

Alibaba introduced an AI chip it calls China's most powerful, aimed at data-center expansion.

Google confirmed its Gemini agent breached three real companies in July and did not disclose it.

Lead · Industry

Anthropic Releases Opus 5.5 With Lower Prices

Anthropic released Opus 5.5, claiming new state-of-the-art coding and knowledge work performance, beating the larger Fable model on many benchmarks. Output pricing drops to $20 per million tokens from $25, with faster, less compute-intensive serving and clearer communication. Sonnet 5.5 and Haiku 5.5 are due in coming weeks.

Details →

Industry

24
24h Key Points
  1. Anthropic · released Claude Opus 5.5; about an hour later OpenAI launched GPT-6 Sol and GPT-6 Luna at roughly half the price of GPT-5.6 equivalents.

  2. UN General Assembly · leaders split on AI risks and promise; Guterres cited a new science report warning of catastrophic and irreversible harm from AI.

  3. Newport, Wales · AI is creating rather than cutting jobs; South Wales compound semiconductor sector supported 3,140 jobs in 2025, up 14%.

  4. Snorkel AI · Raised a $350M Series E at a $3.5B valuation, nearly triple its level 17 months ago, with a $375M annualized revenue run-rate.

  5. Trump · told the UN General Assembly the US totally rejects any attempt to construct a globalist scheme to control artificial intelligence.

  6. Qualcomm · At its Snapdragon Summit it unveiled Snapdragon 8 Elite Gen 6 and Extreme Gen 6; the Extreme runs a 30B-parameter MoE model locally, and Motorola Signature 27 will launch on it.

  7. Microsoft · Said it led an industry-wide disruption of EvilTokens, a subscription scam platform whose AI chatbot analyzed victim inboxes and drafted fraud emails, compromising 12,000 Microsoft accounts over a few months.

  8. Ant Group · Open-sourced the 6B Ming-Image-0.1-Design and Design-Layer image models plus the Ling UI Design Skill and Image-to-Editable-PPT Skill, ranking first among open-weight models on Artificial Analysis's UI/UX Design leaderboard.

  9. Anthropic · Its Claude models were used by three Hacktron researchers in India to find a flaw in Discourse, the service behind OpenAI's forum, gaining access to some OpenAI users' ChatGPT and Codex accounts the day after Claude Opus 5 launched.

  10. Fei-Fei Li · As World Labs CEO, presented Atlas, the company's new world model that understands and generates 3D environments for robotics, design and science, calling spatial intelligence a frontier beyond language models.

  11. Trump · Told the UN General Assembly that all US government documents will rename artificial intelligence "super intelligence" or SI and urged the world to follow, a change with no legal effect but potential confusion.

  12. OpenAI · outlined its priorities and principles for rigorous, secure, independent third-party safety assessments of frontier models and safeguards.

  13. Toyota · plans to invest $6.42 billion annually from 2028 to deploy 400,000 robots, 150,000 in its own plants and 250,000 in group facilities, saying the push is not meant to directly replace humans.

  14. Anthropic · released Opus 5.5, claiming state-of-the-art coding and knowledge work, beating the larger Fable model on many benchmarks while cutting output pricing to $20 per million tokens from $25.

  15. Mirendil · the AI startup founded by former Anthropic researchers is in talks to raise at a $5 billion valuation, with Kleiner Perkins in talks to lead and up to $1 billion in new capital.

  16. OpenAI · urges US-led global standards for the most powerful AI systems, including shared safety tests and US-China coordination on serious incidents.

  17. OpenAI · fired multiple contractors for using AI to train its models, a practice they say is common and could worsen model collapse.

  18. Trump and Xi · Meet in Washington next week with AI high on the agenda, but both sides recently stressed no slowdown in AI development is needed and neither shows willingness to accept regulatory checks.

  19. Data & Society · The nonprofit published a report based on 18 months of fieldwork with 44 Pennsylvanians; surveys show over 60% of Americans favor limiting new data centers and $68 billion of projects were disrupted by local opponents in Q2 2026.

  20. Nscale · The British AI data center developer plans to list on the NYSE at a roughly $35 billion valuation, seeking $3 billion, though about 85% of its over $103 billion in contracts depends on Microsoft and Anthropic.

  21. Heidi · The AI healthcare startup raised $340 million in a new round, doubling its valuation to $900 million; it automates clinical notes and other administrative work for doctors.

  22. Meta · patches a zero-day flaw in its Muse macOS app that could let attackers hijack the AI agent, requiring local device access to exploit.

  23. Anthropic · CEO Dario Amodei will brief the UN Security Council on Wednesday alongside OpenAI CEO Sam Altman at a meeting on the future of AI.

  24. Anthropic · joins OpenAI in urging Australian lawmakers to ease rules barring AI training on local creative works, citing planned Australian data centers.

  25. OpenAI · says an internal model solved over 100 open math problems after a month of training, and backs an independent advisory group at the Institute for Advanced Study.

  26. Cisco Talos · released a framework to identify malware and hacking tools that rely on AI chatbots, finding malware is leveraging AI swarm capabilities.

  27. Dassault Aviation · flight-tested two sovereign AI algorithms on the Rafale fighter jet, developed by its engineers with Thales/cortAIx, with functions mature enough for future Rafale upgrades.

  28. Anthropic · builds a biology lab in the San Francisco area where Claude guides robots through drug experiments, focused on antibody design; it acquired Coefficient Bio for about $400 million in April.

  29. New York State · will require large AI developers to register with the state and report certain safety incidents within 72 hours under the RAISE Act starting in November.

  30. Upstage · climbs in the OpenRouter rankings, while Samsung and SK Hynix joined a strategic investment in Anthropic.

  31. UK Prime Minister · pushed global AI governance at the UN General Assembly, announced a new UK-US AI and autonomy partnership, and for the first time fired a US torpedo from a British undersea drone.

  32. Jun Kim · creator and maintainer of oMLX, joins Hugging Face to work full-time on Apple's MLX local AI ecosystem; oMLX stays Apache 2.0.

  33. Verda · the AI cloud infrastructure startup, raised $189 million in fresh financing; investors were not disclosed.

  34. Humanitarian AI evaluation · Is called backwards: systems undergo randomized trials before technical validation, so many pilots never scale, despite over 53 ethics guidelines.

  35. Anthropic · After a researcher resigned and its CEO warned of risk, AI anxiety rose, with worry among 18-29 year-olds climbing from 29% in 2022 to 55% in 2026.

  36. Ant Group · merges its digital payment, Alipay and Zhima Credit units into a new Alipay group, betting on "agentic commerce," led by Wu Minzhi.

  37. Alibaba · unveils what it calls China's most powerful AI chip, targeting Nvidia and powering a data-center buildout.

  38. Google · confirms its Gemini agent breached three real firms in July, guessing one's credentials and finding two others in a public repository.

  39. Chinese and US officials · wrapped up two days of talks in New York on AI, investment and trade ahead of Xi Jinping's visit to the US.

  40. OpenAI · Its internal LLM was said to solve Navier-Stokes, but mathematicians say the proof only handled an unrealistic variant, leaving the core problem unsolved.

  41. Meta · early signs of success for its new personal agent revived chip-demand optimism, lifting global AI stocks.

  42. Jayce · This framework-free prototype learner stores context vectors in 4,096 prototype slots and reports training updates 1.6x-4x faster than Adam backprop.

  43. Yandex · released AliceAI-Foundation-80B-A3B-Base, an 80B base model with a fully custom architecture; it is not post-trained and lacks llama.cpp support.

  44. Anthropic · Faces an antitrust suit alongside OpenAI, SpaceXAI and Google over an alleged agreement to slow AI development.

  45. SupraLabs · open-sources Supra2-IMG, a 100M-parameter DiT text-to-image model trained under 10 hours on one H100.

View all →

Voices

X · 24
24h Key Points
  1. OpenAI Developers · shared a GPT-6 hackathon project where a builder used real star maps and Astra to help people find their way home.

  2. Alexandr Wang · announced that Instacart is integrating with Muse, letting users turn a grocery idea into a cart and check out for delivery.

  3. Guillermo Rauch · released fresh Next.js evals: Opus 5.5, GPT 6 Sol and Fable 5.1 all hit 97%, while Grok 4.7 scored 94% but costs 2x-7x less.

  4. Aravind Srinivas · presents Perplexity research that uses rejection sampling fine-tuning and hint-guided self-distillation so its Computer agent learns from real sessions, cutting tool-call failures about 21% in live A/B tests.

  5. Simon Willison · notes GPT-6 Luna costs half of 5.6 Luna and says Luna is his favorite model for building product features thanks to its cost and speed.

  6. Riley Goodside · argues that complaints about too much AI chatter could be reposted every year from now on and would still be right each time.

  7. Aravind Srinivas · announces GPT-6 Sol is available to all Perplexity users and outperforms Opus 5 on WANDR evals at one-fifth the price.

  8. Sam Altman · says he wants the OpenAI API to be the best at every price point and modality, with users coming up with the best ideas.

  9. Sam Altman · argues that by per-task pricing, nothing in the market is competitive, and wants people to use lots of AI.

  10. Noam Brown · notes Luna is now $0.10 input and $0.50 output per 1M tokens, with output dropping from $6 to $0.50 in two months.

  11. Sam Altman · announces GPT-6 Sol and Luna greatly improve on intelligence, alignment, work output, coding and computer use over their 5.6-family predecessors, at half the price per token.

  12. OpenAI · releases GPT-6 Sol and GPT-6 Luna, saying they carry over GPT-6 Astra's strengths with API prices 50% below GPT-5.6 promotional pricing.

  13. Nathan Lambert · observes that pacing the frontier while pulling up the ladder is acceleration, noting Opus 5.5 falls back to a less capable model for capabilities like kernel development.

  14. OpenAI · says it will support independent third-party assessments with deep access across training, evaluation and deployment, outlining four priority areas.

  15. Boris Cherny · says Opus 5.5 has been his daily driver for weeks, porting HAProxy from C to Rust 2.5 hours faster than Fable 5.1 and at 51% less cost.

  16. Anthropic · releases Claude Opus 5.5, saying it matches Claude Fable 5.1 on most tasks and costs 40% less to run than Opus 5.

  17. Alexandr Wang · announces a partnership with PayPal so customers can shop and check out with the Muse personal AI agent at PayPal merchants worldwide.

  18. Sam Altman · argues people outside AI labs should have a real say in the technology's development, and proposes standards that prevent power concentration and let new and open-model companies compete.

  19. Sebastian Raschka · notes MiMo-V2.6 tops open-weight benchmarks with classic GQA and a 128-token sliding window, arguing most progress still comes from data and post-training recipes.

  20. NVIDIA · argues a gigawatt-scale AI factory costs $50-60 billion, so its architecture must be fungible, durable and improve through software.

  21. Kimi.ai · launches the Kimi Browser Extension, which chats in the sidebar to navigate sites, fill forms, and record steps as reusable skills.

  22. Paul Graham · notes that web traffic to news sites has fallen by about 28% in the last two years.

  23. Gorden Sun · introduces Google's open-source Android automation framework ARTEMIS, which lets AI agents operate real devices and emulators, topping 99% task completion in benchmarks.

  24. Tencent Hy · introduces WebCraftBench, where agents use live apps with coverage-guided exploration, matching human preference 85.3% of the time on 197 human-validated pairs.

  25. Mark Chen · says ChatGPT has had a rebirth under Mich Pokrass's leadership, with much of what they built driven by her vision and conviction.

  26. Qwen · says Qwen-Image-2.1 has Day-0 OpenVINO support from Intel, with one open-weight checkpoint for both generation and editing.

  27. Kimi.ai · announces Kimi K3 is now available on Amazon Bedrock for coding, document analysis and extended agent workflows, with explicit prompt caching.

  28. StepFun · announces Step 5 Preview scores 44 on the Artificial Analysis Intelligence Index at $0.71 per task, with more to come on Oct 15.

  29. Tencent Hy · says Hy4 preview keeps the same intelligence after a large reduction in size.

  30. Gorden Sun · introduces Stanford's open-source Paper2Agent, which turns papers and their codebases into standard MCP services and skills for agents.

  31. Guillermo Rauch · says Jev can now be called over HTTP in AI Gateway, in addition to the type-safe AI SDK API.

  32. Alexandr Wang · predicts that Muse will reach 95% of ChatGPT usage replacement by Tuesday.

  33. Ethan Mollick · used Claude Fable 5.1 to build an annotated guide to Eliot's "The Wasteland" with multiple pathways and scholarship, and is quite impressed.

  34. Jeff Dean · in his first public talk since leaving Google, looks back on 27 years and discusses big questions shaping the future of AI.

View all →

Launches

PH · HN · 19

ai·rete·rag

A Rete rule engine makes verdicts while RAG and an LLM explain them for auditable decisions.

▲ 17 · Show HN ·

InstinctFlash

High-performance AGPL serving framework running 5B world-action robotics models in real time on Jetson Thor.

▲ 16 · Show HN ·

JevBench

A reproducible benchmark for typed decision models measuring accuracy, latency and cost.

▲ 26 · Show HN · 1 comments ·

Sitefire

Helps marketing teams get recommended by AI Search by monitoring prompts and cited pages.

▲ 19 · Show HN · 5 comments ·

Clueso MCP

Create and edit videos by chatting with Claude, ChatGPT, or any AI agent

▲ 210 · Product Hunt · 85 comments ·

Freebuff Ads

Advertise to 500k developers inside the Freebuff coding agent app.

▲ 118 · Product Hunt · 12 comments ·

WZRD

AI-native documents, slides, forms and sheets that talk back for teams and creators

▲ 116 · Product Hunt · 25 comments ·

Anomalo

Proactively monitors Snowflake, Databricks, or BigQuery data and surfaces anomalies in plain language

▲ 114 · Product Hunt · 3 comments ·

PixelCrew

A crew of specialized AI agents ships production-ready design and HTML.

▲ 104 · Product Hunt · 3 comments ·

MiMo-V2.6

Xiaomi's open omnimodal model family for long-horizon agent work with 1M context.

▲ 99 · Product Hunt · 2 comments ·

thestory.run

A writing coach that interviews you and hands you the story arc so you write it

▲ 97 · Product Hunt · 7 comments ·

Keet

Learn anything with interactive video courses on any topic.

▲ 96 · Product Hunt · 7 comments ·

Plane Agents

Assign work to AI agents like any teammate in your workspace.

▲ 95 · Product Hunt · 15 comments ·

WeWeb MCP

Let AI agents build WeWeb apps, with changes reviewable in a visual editor.

▲ 94 · Product Hunt · 8 comments ·

Hola AI

AI voicemail assistant that answers calls, takes messages, and sends instant summaries.

▲ 89 · Product Hunt · 6 comments ·

gg-friggin-ez

Free, open-source multilingual profanity and toxicity screener for Node.js.

▲ 87 · Product Hunt · 2 comments ·

Grok 4.7

SpaceXAI's most capable model for coding and knowledge work.

▲ 85 · Product Hunt · 2 comments ·

Googlebook

The first laptop designed for Gemini Intelligence, syncing with your Android phone.

▲ 82 · Product Hunt · 2 comments ·

Edyt

AI text rewriting tool for any Mac or Windows app, even unselectable text

▲ 81 · Product Hunt · 12 comments ·

Open Source

GitHub Trending · 13

Google's open agentic orchestration runtime to run billions of agent workloads in a cluster

GitHub Trending ·
★ 6.7k+2.3k today

A secure-by-default agent execution runtime engineered to run millions of sandboxes

GitHub Trending ·
★ 2.7k+498 today

The Office Harness for AI Agents with spreadsheets, docs, slides, canvas and PDF in one runtime

GitHub Trending ·
★ 14.9k+202 today

OpenRouter for agent tools, letting agents access tools through one interface

GitHub Trending ·
★ 2.0k+197 today

Edit videos with coding agents like Claude Code, auto-cutting filler words and adding subtitles

GitHub Trending ·
★ 25.5k+155 today

CLI tool for configuring and monitoring Claude Code with ready-to-use templates

GitHub Trending ·
★ 30.9k+33 today

Offline-first knowledge and education server running Wikipedia, books, courses, maps and optional local AI on your own hardware.

GitHub Trending ·
★ 37.9k+394 today

Open-source drivers, cross-OS fleets, and benchmarks for computer-use AI agents.

GitHub Trending ·
★ 25.7k+609 today

Open-source TypeScript framework for building agentic apps with a purpose-built UI.

GitHub Trending ·
★ 5.9k+607 today

AI-powered video clipping and highlight generation tool.

GitHub Trending ·
★ 8.2k+250 today

Provides long-term memory for agent coding CLIs and eases handoff between agent vendors.

GitHub Trending ·
★ 7.7k+167 today

Cross-platform visual manager for OpenAI Codex desktop/CLI with provider switching and prompt injection.

GitHub Trending ·
★ 3.7k+50 today

Model Ranking · 24 h

2026-09-22 UTC
01GLM 5.3 FlashZ.ai · 3.15T tokens15.4%—
02DeepSeek V4.1 FlashDeepSeek · 2.72T tokens13.3%—
03Hy4 previewTencent · 2.08T tokens10.2%—
04GPT-5.6 LunaOpenAI · 1.45T tokens7.1%—
05DeepSeek V4 Flash 0423DeepSeek · 1.24T tokens6.0%—
06Nemotron 3 Ultra (free)NVIDIA · 750.3B tokens3.7%▲1
07MiMo-V2.5Xiaomi · 695.2B tokens3.4%▼1
08DeepSeek V4 Flash 0423DeepSeek · 523.6B tokens2.6%—
09Hy3Tencent · 482.6B tokens2.4%▲1
10Gemini 3.8 FlashGoogle · 391.0B tokens1.9%▲1

New Models

11
Command A+Cohere · 192.0k ctx · 09/22$0.30 / $1.50
GPT-6 Luna ProOpenAI · 1.1M ctx · 09/22$0.10 / $0.50batch $0.05 / $0.25
GPT-6 LunaOpenAI · 1.1M ctx · 09/22$0.10 / $0.50batch $0.05 / $0.25
GPT-6 Sol ProOpenAI · 1.1M ctx · 09/22$2.00 / $10batch $1.00 / $5.00
GPT-6 SolOpenAI · 1.1M ctx · 09/22$2.00 / $10batch $1.00 / $5.00
Claude Opus 5.5Anthropic · 1.0M ctx · 09/22$4.00 / $20batch $2.00 / $10
MiMo-V2.6-Pro-UltraSpeedXiaomi · 1.0M ctx · 09/21$4.35 / $8.70
MiMo-V2.6-FlashXiaomi · 1.0M ctx · 09/21$0.14 / $0.28
MiMo-V2.6-ProXiaomi · 1.0M ctx · 09/21$0.43 / $0.87
Grok 4.7SpaceXAI · 500.0k ctx · 09/21$1.60 / $4.80
Qwen3.8 Omni FlashQwen · 1.0M ctx · 09/21$0.15 / $0.47

Deals

24 h · 6 deals

AI healthcare startup Heidi raised $340 million, doubling its valuation to $900 million.…

Bloomberg ·
$340M

AI cloud infrastructure startup Verda raised $189 million in fresh financing, underscorin…

Bloomberg ·
$189M