Skip to content

#其他

0 today

Jul 10Friday

AI HOT (Curated Pool)

Meta launches Muse Spark 1.1, an agentic model that punches near flagship level on agent tasks at a very low price

Meta released Muse Spark 1.1 via a new API, built around delegating tasks to parallel sub-agents and cross-device GUI control. It leads on 4 agent benchmarks—JobBench jumped 3.2× from 17.0 to 54.7. Coding trails flagships: Terminal-Bench 80.0 vs GPT 5.5's 83.4, SWE-Bench Pro 61.5 vs Opus 4.8's 69.2. Zuckerberg pitched it as very low price, aiming for strong-enough agent performance with cheap-enough coding. The post doesn't disclose exact pricing or rollout scope.

Why it matters: Meta's new Model API opens Muse Spark 1.1, which splits complex tasks across parallel sub-agents and controls GUIs across devices.

Financial Times · Technology

SK Hynix raises $26.5bn in Nasdaq IPO

SK Hynix debuted on Nasdaq, raising $26.5bn in the year's largest global IPO. The funds are earmarked for HBM capacity expansion, feeding Nvidia and other AI chipmakers. The post doesn't disclose the offering price or first-day move, but the size alone shows the market is still doubling down on AI compute supply chains.

Why it matters: SK Hynix raised $26.5bn in its Nasdaq debut, with the money going to HBM expansion that supplies AI chipmakers like NVIDIA, a window into where money is flowing in the AI compute supply chain.

TechCrunch · AI

OpenAI says GPT 5.6 is the ‘preferred model’ for Microsoft Copilot 365 amid breakup chatter

At the GPT 5.6 launch, OpenAI announced the model will be the 'preferred model' for Microsoft 365 Copilot across Word, Excel, PowerPoint, and Cowork. This is a direct response to Bloomberg's earlier report that Microsoft is swapping some OpenAI software for its own MAI models to cut costs. The post doesn't spell out what 'preferred model' actually means. And the earlier report never said ChatGPT would stop powering Microsoft apps—just that Microsoft is leaning more on its own tech. This reads like a PR move that doesn't contradict the cost-cutting story.

Why it matters: OpenAI says GPT 5.6 is the preferred model for Microsoft 365 Copilot, while Bloomberg reported Microsoft is swapping in its own MAI for some OpenAI software, two signals worth reading together.

The Verge · AI

Microsoft's carbon emissions rose 25% last year as AI growth outpaces sustainability fixes

Microsoft's 2026 sustainability report shows 2025 carbon emissions are 25% above its 2020 baseline, driven by data center construction and supply chain. The company admits current sustainability solutions aren't keeping pace with AI demand, making its 2030 carbon-negative goal harder to reach. The report doesn't lay out a concrete catch-up plan.

Why it matters: Microsoft's 2026 sustainability report discloses 2025 carbon emissions 25% above its 2020 baseline and concedes current plans cannot keep up with AI demand, putting its 2030 carbon-negative goal under pressure.

Computing Life · Share · Yage

The chat box illusion: why AI agents need email-style interfaces, not chat

This piece argues that chat-box interfaces in tools like Cursor and Claude Code nudge users toward vague prompts and instant replies, robbing AI agents of the quiet time needed to compile, run tests, and self-correct. The author proposes replacing turn-by-turn chat with email-style async workflows: send a detailed task brief with attachments and local paths, then close the window and review the result later. It names Manus's email task entry and the startup AgenticMail as early examples. The post does not disclose latency or success-rate data for these email-based agent products.

Why it matters: Argues that Cursor and Claude Code chat boxes push vague prompts and instant replies, squeezing out the time AI needs to compile, test and fix itself.

Computing Life · Share · Yage

Cloudflare moves machine-traffic payments to the edge gateway

Cloudflare launched Monetization Gateway on July 1, 2026, letting sites, APIs, and MCP tools charge per machine request. It returns HTTP 402 with a price quote; agents pay via the x402 protocol using stablecoins, with verification and settlement handled at Cloudflare's edge. This expands beyond the earlier Pay Per Crawl product to cover any machine request. The post notes x402 handles pricing and payment but not authorization, disputes, or bot detection—those fall to Google AP2, Mastercard verifiable intent, and Visa's Trusted Agent Protocol. Real-time data, specialized databases, and high-trust execution tools are most likely to attract agent payments; generic web content is hard to monetize. The real test is whether mainstream AI SDKs and agent tools adopt x402 budget options by default, and whether any resource owners publicly disclose actual revenue.

Why it matters: Cloudflare launched Monetization Gateway, letting websites, APIs and MCP tools charge for every machine request, a shift in how machine traffic moves from free scraping to something that can be priced.

TechCrunch · AI

OpenAI's No. 2 Fidji Simo steps down from full-time role

Fidji Simo is stepping down to a part-time advisory role after her medical leave for a neuroimmune relapse proved longer than expected. She joined as CEO of Applications in May 2025, consolidating business and product under her. The gap hits as OpenAI eyes a possible IPO and races to catch Anthropic in enterprise. CPO Kevin Weil and CMO Kate Rouch also recently left, deepening the leadership churn.

Why it matters: OpenAI's business and product lead steps back to part-time adviser as her recovery from a neuroimmune condition runs slower than expected, leaving a key management seat in flux.

The Verge · AI

Fidji Simo steps down from leading OpenAI's AGI work due to illness

Fidji Simo is stepping down from leading OpenAI's AGI readiness team due to illness, transitioning to a part-time advisor role. She had been on medical leave for a few months. Simo, who is also CEO of Instacart, joined OpenAI's board in September 2024 and later took over AGI preparedness. OpenAI hasn't named a successor or detailed how this affects the AGI readiness timeline.

Why it matters: OpenAI's AGI preparation lead steps back for health reasons, with no successor or timeline named yet.

TechCrunch · AI

OpenAI launches GPT-5.6 family, pushing coding efficiency and cybersecurity

OpenAI dropped GPT-5.6 in three tiers: Sol (workhorse), Terra (mid-range), and Luna (budget). Sam Altman told CNBC Sol is 54% more token-efficient on coding tasks. The company calls it their strongest cybersecurity model yet, covering threat modeling, code review, and blue teaming. The Trump administration previously tried to restrict its rollout over misuse fears. ChatGPT Work, an enterprise companion tool, also launched. The post doesn't disclose pricing or availability dates.

Why it matters: OpenAI shipped three model tiers at once, with flagship Sol cutting token use 54% on coding and adding threat modeling and code review.

TechCrunch · AI

Lyzr used its own AI agent to run its $100M Series B fundraise

Lyzr, a startup that builds enterprise AI agents, let its own agent SivaClaw lead its $100M Series B at a ~$500M valuation. The agent fielded questions from 130+ investors, drafted memos, and tracked which slides held attention. Lyzr says it pulled in $400M in interest without the founders ever flying out for traditional roadshows. The round doubled as a product demo, and the bigger story is how much capital is chasing AI deals right now.

Why it matters: Lyzr let its own agent SivaClaw run its $100M Series B, with the agent answering questions from over 130 investors and drafting the investment memo, so the raise doubles as a product demo.

AI HOT (Curated Pool)

Can AI answer the $3 trillion question?

Sequoia partner David Cahn estimates 2026 AI infrastructure spending at $1.5 trillion, meaning the industry must generate $3 trillion in revenue to justify the investment. Rising memory costs and exotic chips may push that number higher. On the revenue side, Anthropic is at roughly $60B ARR and OpenAI earned $13B in 2025 — a large gap remains.

Why it matters: Sequoia's David Cahn puts 2026 AI infrastructure spend at $1.5 trillion, needing $3 trillion in revenue to break even.

AI HOT (Curated Pool)

Bun rewrites from Zig to Rust to fix memory safety bugs

Jarred Sumner announced Bun is being rewritten from Zig to Rust. The trigger was a long list of use-after-free, double-free, and memory leak fixes in v1.3.14—mixing GC with manual memory management proved too error-prone. With 22M+ monthly downloads and adoption by tools like Claude Code, the team decided one-off bug fixes aren't sustainable. A pre-release Claude Fable 5 assisted the rewrite. The post does not disclose a migration timeline.

Why it matters: Bun creator Jarred Sumner announced a full rewrite from Zig to Rust after a batch of memory crashes and leaks fixed in v1.3.14 traced back to mixing garbage collection with manual memory management.

AI HOT (Curated Pool)

Microsoft releases Flint, a visualization language that lets AI agents generate charts from compact JSON specs

Flint is a Microsoft Research project that gives AI agents a compact way to produce polished charts. You supply data, semantic types (e.g., YearMonth, Profit), a chart type, and encodings—the compiler then fills in scales, axes, layout, and color schemes automatically. It supports 46 chart types and can render to Vega-Lite, ECharts, or Chart.js. An MCP server is included so agents can call Flint directly in workflows. The project is open source, available via npm, and has an online editor.

Why it matters: Microsoft Research's open-source Flint lets an AI agent supply only data and column semantics, and the compiler fills in ticks, axes and colors on its own.

Hacker News front page

Ello breaks down the architecture of a real-time AI tutor that responds to a 5-year-old in under one second

Ello built an AI tutor for kids ages 4-9 that teaches math and reading. Standard agent tool loops were too slow—frontier models take 2–3 seconds for the first token, and kids tuned out. Their fix: the model streams multiple actions in a single response, and an interpreter executes each action as it arrives, so the child waits for only the first ~30 tokens. They also split the tutor into two agents—a real-time converser and an async planner that updates strategy during gaps when the child is thinking or speaking. Both agents share an append-only event log to avoid coordination delays. The post does not disclose end-to-end latency numbers or which models are used.

Why it matters: Ello's AI tutor for 4-9 year olds swaps the standard agent loop for a single streaming response with multiple actions, so kids see a reaction after the first 30 tokens.

The Verge · AI

OpenAI's ChatGPT browser Atlas is shutting down less than a year after launch

OpenAI launched the Atlas browser in October 2025, pitching it as a way to browse, fill forms, and book tickets with ChatGPT. The company now says it will shut down on August 31, 2026 — less than a year after launch. The post does not disclose the reason or any user numbers. A browser killed this fast usually means low adoption or a strategic pivot.

Why it matters: OpenAI's Atlas browser ran from October 2025 to August 31, 2026, and the shutdown notice gives no reason and no user numbers.

MIT Technology Review · AI

Anthropic found a hidden space where Claude puzzles over concepts

Anthropic built a tool called the Jacobian lens (J-lens) and used it to uncover a hidden region—dubbed J-space—inside Claude Opus 4.6. J-space surfaces words related to what the model is about to say, but those words may not appear in the final output. Anthropic claims monitoring these words offers a new way to understand and control its models. The company published a paper and released a public demo with Neuronpedia. Goodfire chief scientist Tom McGrath called the work “very good and interesting.” The post does not disclose J-lens false-positive rates or any impact on model performance.

Why it matters: Anthropic's J-lens tool found a J-space in Claude Opus 4.6's middle layers where concept words surface before the answer, a path into watching a model's internal activity.

TechCrunch · AI

New York Times says OpenAI hid evidence in ChatGPT copyright trial

The New York Times and The Daily News filed a sanctions motion, accusing OpenAI of lying about its inability to search training data and chat logs for copyrighted material. OpenAI had claimed such searches were technically infeasible, but the publishers say internal tools and datasets exist that can do exactly that. The court has not ruled yet; OpenAI's response is not in the article.

Why it matters: The New York Times and Daily News say OpenAI has an internal tool for searching training data, which cuts against its claim that providing the data was technically impossible.

TechCrunch · AI

Paris voice AI startup Gradium raises $100M seed round, backed by Nvidia

Gradium builds ultra-low-latency voice models so AI conversations feel instant, without awkward pauses. It first came out of stealth last December with $70M, then reopened the seed round to bring in Nvidia and others, hitting $100M total. The cash goes toward opening a Bay Area office and competing for talent near Anthropic and OpenAI—a telling move for a Paris-based startup. It already counts Renault as a customer, but faces stiff competition from ElevenLabs (valued at $11B) and Google's Gemini voice capabilities.

Why it matters: Gradium left stealth in December with $70M and has now grown its seed round to $100M with Nvidia joining, another data point for capital flowing into low-latency voice models.

Hacker News front page

GLM 5.2 prepares a UK SME's quarterly VAT return at near-human accuracy for under 1% of the cost

Vineyard Finance tested GLM 5.2 on a real quarter of UK VAT bookkeeping: 59 transactions processed in 68 minutes at a raw token cost of $2.73. The final VAT return's Box 5 was off by only 7 pence. The model operated accounting software via CLI and was scored on 6 criteria per transaction; it failed 20 of 354 checks. The one serious error was misclassifying founder shares as 'Capital Account' instead of 'Unpaid Shares', which carries legal risk. The other 19 mistakes had no financial impact but are errors a skilled accountant wouldn't make. The model had internet access and two user notes; no cheating was detected. The post does not disclose the exact quantization, but believes it is FP16 or FP8.

Why it matters: Vineyard Finance ran a full quarter of UK VAT filing on GLM 5.2 alone: 59 transactions, 354 checkpoints with 20 flagged, a 7-pence Box 5 gap, and $2.73 in token cost.

TechCrunch · AI

How did the US government decide OpenAI's frontier model Sol was safe to release?

OpenAI is rolling out Sol, a frontier model on par with Anthropic's Fable, which the White House briefly banned. Mina Narayanan of Georgetown's CSET says she has no visibility into the government's review process. Anthropic mentioned building a jailbreak classifier and defense-in-depth, but the actual dialogue between the government and the labs remains opaque.

Why it matters: OpenAI released Sol to the public with Fable-level capability, yet how either model cleared government safety review is still unexplained, even to Georgetown researchers.

AI HOT (Curated Pool)

OpenAI launches ChatGPT Work desktop app, integrating Codex and GPT-5.6

OpenAI combined Codex and ChatGPT into a single desktop app called ChatGPT Work. Powered by Codex and GPT-5.6, it can work across apps and files, running complex projects for hours. It also includes new coding workflows, a Chrome extension, an improved built-in browser, and faster Computer Use driven by GPT-5.6. The post doesn't disclose launch date, pricing, or system requirements.

Why it matters: OpenAI folded Codex and GPT-5.6 into the ChatGPT Work desktop agent, which runs multi-step projects across apps and files.

AI HOT (Curated Pool)

Anthropic launches 'Hard Questions' initiative, inviting the public to ask tough questions about AI

Anthropic opened a public page today called 'Hard Questions,' explicitly asking people to submit their toughest concerns and hopes about AI. This isn't a one-off PR move—they simultaneously released findings from surveys of 52,000 Americans and 81,000 Claude users across 159 countries, plus multiple in-person focus groups, as a baseline for understanding public sentiment. Anthropic commits to publicly tracking what actions they take in response and where they fall short. The post doesn't specify a response timeline or evaluation criteria; I'd treat this as a transparency experiment and wait to judge the quality of follow-up reports.

Why it matters: Anthropic opened a hard questions page to collect public worries and hopes about AI, alongside a survey of 52,000 Americans and interviews with 81,000 Claude users in 159 countries.

Hacker News front page

A browser tool that reads a model's internal concepts layer by layer before it speaks, using the Jacobian lens

Lucid lets you type a prompt in the browser and watch which concepts activate inside small models like Qwen and Pythia before they answer, layer by layer. It uses a Jacobian lens that costs a single forward pass, no account or install needed. The author borrows Anthropic's J-space framing, treating the model's reportable internal representations as a global workspace. Zener cards serve as a demo: the lens reads the answer four layers before the model speaks. Currently only 0.5B–3B open models are supported; the post doesn't say when larger models will be added.

Why it matters: The author turned the Jacobian lens into a browser tool: type a sentence and step through which concepts small models like Qwen and Pythia activate before answering, no signup needed.

TechCrunch · AI

Nvidia's stock falls to pre-AI-boom levels as the compute market it built turns on it

Nvidia's stock dropped 15% from its May peak, now cheaper than the S&P 500 average on a forward earnings basis. Money is still pouring into AI infra, but mostly into memory companies — Micron nearly tripled over the same period. The GPU shortage has eased; memory is now the data center bottleneck. The post cites Bloomberg for details but doesn't include revenue figures itself.

Why it matters: Nvidia shares are down 15% from their May high and now trade below the average S&P 500 valuation, while Micron's market cap has nearly tripled as compute money shifts from GPUs to memory.

The Verge · AI

OpenAI releases GPT-5.6 and announces ChatGPT Work for enterprises

OpenAI launched GPT-5.6 after receiving government approval, ending a months-long limited preview. They also announced ChatGPT Work, an enterprise-focused version, though the post doesn't detail its features, pricing, or launch date. Specific benchmarks or performance gains for GPT-5.6 aren't covered either — only the release and product names are confirmed.

Why it matters: GPT-5.6 moves from limited preview to general availability after government clearance, and OpenAI names its enterprise product ChatGPT Work.

Financial Times · Technology

Microsoft’s early AI lead has become a test of faith

The FT argues Microsoft’s Copilot strategy, built on its OpenAI tie-up, hasn’t yet translated into clear revenue gains. Azure growth is slowing, enterprise willingness to pay for Copilot remains uncertain, and in-house model efforts lag. The AI premium the market gave Microsoft now hinges on hard financial delivery.

Why it matters: This FT column traces how Microsoft's OpenAI tie-up gave it a Copilot head start that has yet to turn into revenue, with slower Azure growth and lagging in-house models adding pressure.

AI HOT (Curated Pool)

Anthropic's Long-Term Benefit Trust appoints Ben Bernanke as trustee

Anthropic's Long-Term Benefit Trust added former Fed Chair Ben Bernanke as a trustee. The LTBT is an independent body that checks whether the company stays true to its public-benefit mission. Bernanke, a Nobel laureate who steered the US through the 2008 financial crisis, will advise on how AI affects workforces and economies. Trustees hold no equity, share no profits, and are paid only for their time.

Why it matters: Former Fed chair Ben Bernanke joins Anthropic's long-term benefit trust to help judge AI's impact on labor and the economy, adding a macro view to the company's long-term governance.

Hacker News front page

AI 2040: Plan A — a roadmap to delay superintelligence, enforce transparency, and replace the race with mutually assured compute destruction

Six authors, including a former OpenAI staffer, propose an international deal to delay superintelligence until 2040 by mandating total AI R&D transparency and letting dozens of companies catch up globally, then locking in a regime of mutually assured compute destruction. They stress this is a recommendation and scenario stress-test, not a prediction. The post argues that current frontier labs acknowledge the risks of loss of control and power concentration yet race ahead anyway, explicitly naming the CEOs of OpenAI, Anthropic, xAI, and Google DeepMind as possibly viewing themselves as the lesser evil. Supplements cover verification, transparency, security, and economic modeling, but the main body does not spell out the political path to getting nations to sign on.

Why it matters: Six authors propose an international agreement to delay superintelligence to 2040, trading the arms race for mandatory transparency and compute checks.

Jul 9Thursday

Hacker News front page

Pangram extension data: 1 in 4 longform social posts is fully AI-generated

Pangram pulled anonymized scan data from its Chrome extension. The overall AI-generation rate was 13.8%, but for posts over 250 words it hit 25.72%. LinkedIn dominated: it accounted for 62% of all AI-flagged content, with more than 40% of longform posts marked fully AI-generated. On X, only 53.2% of articles scanned as fully human-written. Substack fared better but still saw 21.9% of posts flagged. The post doesn't disclose total sample size or the collection window, so treat the absolute numbers with caution, but the platform-level pattern is clear.

Why it matters: Pangram counted AI-generated long posts from anonymous scans by its own Chrome extension, and LinkedIn produced 62% of flagged AI content from a third of the scans.

AI HOT (Curated Pool)

Tesla's Optimus Gen 3 reportedly finalized; Musk demands 2,000–2,500 units/week by year-end or he'll replace the entire procurement team

Per LatePost and supply chain sources, Tesla issued parts procurement guidance for Optimus: ramp to 1,000 units/week by September and 2,000–2,500/week by year-end, implying ~100k units/year capacity. At a late-June exec meeting, Musk approved the final Optimus Gen 3 design and said he'd fire the entire procurement team if the year-end target isn't met. The Fremont factory's former Model S/X line is now the robot line; concrete orders for hundreds of units in August are already placed. Musk himself tempered expectations, saying initial production will be 'extremely slow' because everything is new and the robot involves ~10,000 unique parts.

Why it matters: Tesla has signed off on the final design of Optimus Gen 3 and told suppliers to target 1,000 units a week and 2,000 by year-end, converting the old Model S/X line in Fremont to robots.

Hacker News front page

Frigade reverse-engineers web apps into agent-callable tools

Frigade built a browser agent that signs into web apps, watches their internal API calls, and auto-generates authenticated tool recipes with schemas. LLMs can then use these recipes to act inside the app without custom code, and recipes self-update when APIs change. The post notes GraphQL was the hardest API to standardize. Only demo links and a HN post are available; no pricing or launch date is disclosed.

Why it matters: Frigade's browser agent logs into web apps, watches their internal API calls and writes tool recipes with auth and input/output schemas, updating itself when endpoints change.

Hacker News front page

Ways to think about token pricing — Benedict Evans

Benedict Evans argues token pricing is in a supply crunch that won't last. Over a trillion dollars in data center capex is coming, and inference efficiency is improving fast. The recent demand surge is driven almost entirely by one use case—software development—which is a small market. He leans toward foundation models becoming low-margin commodity infrastructure rather than capturing sustainable pricing power. The piece frames the question around four variables: how many use cases will pay for frontier performance, how long the frontier keeps moving, whether competition stays fierce, and how much value the model itself captures versus the tooling wrapped around it. No specific price forecasts or timelines are given.

Why it matters: Benedict Evans breaks token pricing into supply, demand, inference efficiency and capex, and concludes foundation models may end up a low-margin infrastructure business.

TechCrunch · AI

Anthropic, OpenAI, and SpaceX are bigger than the last 25 years of tech exits

A new Pitchbook report estimates that SpaceX, Anthropic, and OpenAI together will generate more exit value than all U.S. VC-backed exits since 2000. SpaceX already went public at $1.77 trillion; Anthropic and OpenAI are each pushing toward trillion-dollar valuations. The post doesn't give a precise combined figure, but the concentration in AI and space is historic.

Why it matters: A historical yardstick for the AI valuation debate: three exits would top every US VC exit since 2000, with SpaceX already listed at $1.77 trillion.

Hacker News front page

Meta launches Muse Spark 1.1, a multimodal reasoning model for agentic tasks

Meta Superintelligence Labs released Muse Spark 1.1, a multimodal reasoning model with major gains in tool use, computer use, and coding. It zero-shot generalizes to new tools and MCP servers, manages a 1M-token context window, and compacts memory to keep critical steps. The model orchestrates multi-agent systems, delegating tasks to parallel subagents to cut end-to-end latency. Coding improvements cover bug fixes, feature additions, and large code migrations in complex codebases. It is live in Meta AI's Thinking mode and in the new Meta Model API public preview.

Why it matters: Meta Superintelligence Labs released Muse Spark 1.1, which calls unfamiliar tools and MCP servers without extra training and supports a 1M-token context with active memory compression.

Hacker News front page

Anthropic adds a usage reflection dashboard to Claude

Anthropic launched a beta feature called Reflect inside Claude’s web and desktop settings. It visualizes your chat activity over the past 1–12 months: when you use Claude most, which topics dominate, and what task patterns emerge. The report also maps your usage to Anthropic’s 4D AI Fluency Framework—Delegation, Description, Discernment, Diligence—and offers practical tips, like starting a Project instead of re-explaining context. Incognito chats, health-integration conversations, and source files from connected tools are excluded; sensitive topics appear only at a high level. Available now for Free, Pro, and Max users with Memory turned on; Cowork conversation support is coming soon.

Why it matters: Anthropic is testing a Reflect feature in Claude on web and desktop that turns 1 to 12 months of chat history into a visual view of usage times, frequent topics and main uses.

Ben's Bites

SpaceXAI and Cursor trained Grok 4.5, a model 6x cheaper than Opus

SpaceXAI and Cursor jointly trained Grok 4.5, landing between Opus 4.7 and 4.8 in performance but 6x cheaper than Opus and 3x cheaper than GPT-5.5 on a per-token basis. OpenAI rolled out GPT-5.6 (Sol, Terra, Luna) to all users; early testers say Sol is less smart than Fable but far more reliable. ChatGPT Voice got new GPT-Live-1 and Live-1-mini models that can talk while you speak and use GPT-5.5 in the background. Anthropic extended Fable 5 access for Claude subscribers to July 12—the post doesn't explain the repeated delays. Meta introduced Muse Image and Muse Video; image editing and text rendering look solid, but images still have an AI look, and the video model is in preview.

Why it matters: Grok 4.5, trained jointly by SpaceXAI and Cursor, is live in Cursor with an open API, sits between Opus 4.7 and 4.8 in performance, and costs 6x less per token than Opus.

TechCrunch · AI

Ollama raises $65M Series B, reaches nearly 9M users

Ollama, the tool that lets devs run open-weight models locally on their PCs, raised a $65M Series B led by Theory Ventures. That follows a $15M Series A led by Benchmark, bringing total funding to $88M. Launched in 2023, it has 176K GitHub stars, nearly 17K forks, and close to 9M users. The post doesn't disclose valuation, revenue, or commercialization plans.

Why it matters: Ollama raised a $65M Series B, $88M across two rounds, and has close to 9M users, showing running open models on a personal computer has reached real scale.

OpenAI News

OpenAI turns its bio bug bounty into an ongoing program, doubling rewards to $50K starting with GPT-5.6

OpenAI is turning its GPT-5.5 Bio Bug Bounty into an ongoing private program, now called the OpenAI Bio Bounty Program. The focus stays on universal jailbreaks that beat its biosafety challenges. Rewards jump from $25,000 to $50,000 for both GPT-5.6 and GPT-5.5, with smaller payouts possible for partial wins. GPT-5.5 testing ends July 27, 2026; after that only GPT-5.6 is in scope. Applicants need a ChatGPT account, must sign an NDA, and past GPT-5.5 applicants don't need to reapply.

Why it matters: OpenAI upgraded its bio-safety bounty from a one-off to a permanent program with doubled rewards and clearer rules — a substantive safety-mechanism update.

AI HOT (Curated Pool)

OpenAI launches GPT-5.6 family: Sol, Terra, Luna, pushing performance per dollar

OpenAI released the GPT-5.6 family on July 9: flagship Sol, balanced Terra, and low-cost Luna. Sol scores 53.6 on Agents' Last Exam, beating Claude Fable 5 by 13.1 points at roughly one-quarter the estimated cost. A new `ultra` mode coordinates parallel agents to cut latency and lift scores on BrowseComp and Terminal-Bench 2.1. Sol also tops the Artificial Analysis Coding Agent Index at 80, using less than half the output tokens of Fable 5. Terra and Luna outperform Fable 5 at about one-sixteenth the cost. OpenAI ran extensive red-teaming and automated testing, and hardened safeguards with external partners during a preview period.

Why it matters: OpenAI shipped three GPT-5.6 models at once, and the flagship Sol beats Claude Fable 5 by 13.1 points on long-horizon agent tests at about a quarter of the cost.

AI HOT (Curated Pool)

OpenAI launches ChatGPT Work, an agent that acts across apps and stays with projects for hours

ChatGPT Work is an agent that acts across apps and files, powered by the new GPT‑5.6 model. It breaks complex projects into steps, creates slides, sheets, docs, and web apps, and can run scheduled tasks while you're away. Nearly all teams inside OpenAI use it; early external users include Zapier, RingCentral, Virgin Atlantic, and NVIDIA. The post does not disclose pricing details, only that it's available starting today.

Why it matters: ChatGPT Work and GPT-5.6 let agents act across apps and files, break projects into steps, and pause for user approval.