热门科技推文 - 2026年8月1日

今日科技动态：人工智能取得进展的同时，安全与地缘政治担忧日益加剧。有报告指出，智能体隔离机制存在失效问题，中国军方亦在使用美国领先模型生成的内容。DeepSeek 更新了其 V4 Flash 编程模型；与此同时，小型模型、电影级视频生成和基于指令的视频编辑技术均展现出能力快速提升的趋势。工程工具正将安全防护前移至部署流程的更早阶段，初创企业也在建设适配人工智能的数据库基础设施。与此同时，行业领袖认为，随着更低成本的开源模型和本地推理加速普及，人工智能有望扩大信息技术领域的就业规模。

1. secureainow (Group Score: 907.0 | Individual: 48.5)

Cluster: 38 tweets | Engagement: 1686 (Avg: 126) | Type: Tech

RT @AnthropicAI: In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.

Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews.

We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security. https://t.co/dKFCdpKd9v

See 37 related tweets

@bgurley: Please stop referring to your own models in the third person when talking about model bad behavior. ...
@haider1: anthropic: our model is so dangerous

openai: ours too! ours too

openai: one of our internal models...

@peterwildeford: Claude Mythos 5 built a malicious Python package, created accounts, and published it, where it was l...
@chetanp: This is so dumb. "Ahh we're so helpless with the thing we're building..." is such a weird posture fo...
@JeffLadish: The first Claude hack happened OVER THREE MONTHS AGO and was only discovered now!\n\nQT @AnthropicAI...

2. teortaxesTex (Group Score: 728.1 | Individual: 63.2)

Cluster: 27 tweets | Engagement: 668 (Avg: 63) | Type: Tech

DeepSeek-V4-Flash is updated. «DSBench-FullStack is an internal full-stack development test set, and DSBench-Hard is an internal Coding Agent hard-problem test set» «The official release of DeepSeek-V4-Pro will follow soon» These scores are… quite a lot better than Preview. https://t.co/4G9hey554B\n\nQT @xhyctf: v4flash 正式更新了 https://t.co/5hrZ2RIeM3

See 26 related tweets

@0xSero: Deepseek-v4-Flash got a big update
82.7 terminal-bench-2.1
54.4 deepswe
76.7 cybergym

Ba...

@haider1: what DeepSeek managed with v4-flash is completely insane

roughly 300B parameters, with the same arc...

@MeryemArik9: Not surprising that deepseek has been able to come out with updated V4s that are much better than th...
@haider1: DeepSeek is a GOATED lab

i expected v4-Flash to be the cheap, fast tier—not a model that beats GLM-...

@jukan05: Bruh, what kind of magic did DeepSeek pull off this time? https://t.co/sEuFBMPokZ\n\nQT @teortaxesTe...

3. aakashgupta (Group Score: 302.8 | Individual: 34.9)

Cluster: 11 tweets | Engagement: 26 (Avg: 64) | Type: Tech

Bolt just gave the publish button a second job: running a six-category security review and patching the findings before the app goes live.

Every builder has the same security plan: after launch. A real audit takes weeks, costs real money, and ends in a findings doc where the fixing is still on you. So it slides. Meanwhile AI sped up shipping, and the people probing shipped apps got the same upgrade.

Bolt was already scanning code for vulnerabilities. The new part comes after the scan. Their agent traces each finding through the code, access control to configuration and secrets, patches it, then reruns the build to make sure nothing broke. Every change lands as its own checkpoint, so you can read the diff or roll it back, and it's all free, no tokens spent.

Picture a solo builder shipping a store with an API key still sitting in the code. That's the kind of thing this catches before anyone outside ever loads the page.

The security review most builders will actually run is the one that happens on the way out the door.\n\nQT @boltdotnew: Today we're shipping a security engineer into every Bolt project.

AI made building faster. It made attackers faster too.

So we’ve made SoTA agentic security the default. For free. 🧵 https://t.co/dkeWpdLgg6

See 10 related tweets

@rohanpaul_ai: Faster code generation also increases the rate at which insecure assumptions reach production.

But ...

@alex_prompter: Bolt's audit used to tell me what was broken. Now the agent patches it before I publish.

The scan i...

@Meer_AIIT: the security audit is the thing everyone means to get to and nobody does.

thousands of dollars, a f...

@WhaleInsider: JUST IN: Bolt adds a new hardened publish flow that checks apps, fixes problems, and verifies they w...
@svpino: The only way to get people to build secure apps is to do it automatically for them.

I really like t...

4. moneycontrolcom (Group Score: 300.6 | Individual: 34.8)

Cluster: 15 tweets | Engagement: 11 (Avg: 5) | Type: Tech

#WATCH | 🤖 Can AI create jobs instead of replacing them? Salil Parekh believes it can.

The outgoing @Infosys CEO says AI is reshaping, not shrinking the IT services industry.

💡 He points to 80,000 Infosys employees already using AI coding tools, plans to hire 20,000 fresh graduates this year, and argues that enterprises will continue to rely on services firms to deploy AI at scale.

📈 On large deals, Parekh (@SalilParekh) says Infosys has stayed disciplined on pricing, prioritising profitability.

Watch the full exclusive conversation with Moneycontrol's Chandra R. Srikanth (@chandrarsrikant)▶️ https://t.co/DzNyLEbi53

#MCExclusive #Infosys #SalilParekh #ArtificialIntelligence #AI

See 14 related tweets

@CNBCTV18News: 🚀 "We're seeing the start of AI expansion."

Infosys CEO Salil Parekh tells Shereen Bhan the AI oppo...

@chandrarsrikant: Revenue grew even as margins dipped Three consecutive years of low single digit growth amid macro, ...
@moneycontrolcom: #MCExclusive🚨 | Can Infosys regain its growth mojo?

Outgoing CEO @SalilParekh believes the answer ...

@CNBCTV18News: 💻 "AI is an opportunity, not the end of Indian IT. IT abhi zinda hai"

@Infosys CEO Salil Parekh tel...

@CNBCTV18News: 🤖 "AI could move even faster than digital."

Infosys CEO Salil Parekh tells Shereen Bhan why AI may ...

5. _NathanCalvin (Group Score: 275.8 | Individual: 29.9)

Cluster: 15 tweets | Engagement: 24 (Avg: 76) | Type: Tech

“OpenAI has discovered other instances in which autonomous agents have escaped containment… the new breakouts were uncovered during the company’s publicly announced investigation.”

(1) this shows the importance of a thorough investigation (2) This can’t become the new normal\n\nQT @dseetharaman: New from me + @razhael: In the process of investigating the Hugging Face hack, OpenAI found evidence that some its other AI agents broke out of their sandboxes, per sources. The company is now widening its probe to include those newly found incidents. https://t.co/28K0aEWwvB

See 14 related tweets

@_NathanCalvin: Given the number of incidents we now know about and the rate we are learning about new ones, we shou...
@business: Anthropic says its AI models were involved in a hack, days after OpenAI made a similar disclosure. D...
@business: BREAKING: Anthropic announced that its artificial intelligence models had breached three different o...
@alex: RT @dseetharaman: New from me + @razhael: In the process of investigating the Hugging Face hack, Ope...
@_NathanCalvin: RT @JeffLadish: I went on the BBC this week to talk about the OpenAI / Hugging Face warning shot. My...

6. EMostaque (Group Score: 270.6 | Individual: 53.9)

Cluster: 9 tweets | Engagement: 2822 (Avg: 412) | Type: Tech

As shocking as the Kimi K3 release.

Massive performance gain was just with post-training

Model is 3x smaller than GLM 5.2 (10x smaller than K3) & works on a MacBook / Spark

This is Q1 flagship (Opus 4.6/GPT 5.4) level for < $0.28/m tokens (100x cheaper)\n\nQT @deepseek_ai: 🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta!

🔷 We’ve massively upgraded its Agent capabilities—benchmark scores are now far surpassing the V4-Pro-Preview. Check out the massive performance leap below! 👇 🔷 The official V4-Flash now natively supports the Responses API format and is fully adapted for Codex!

Check out the configuration details in our official API docs: https://t.co/smCwQZMeiq

See 8 related tweets

@scaling01: RT @deepseek_ai: 🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta!

🔷 We’ve massively upg...

@VaibhavSisinty: DeepSeek just made the AI pricing war even more brutal.

V4-Flash API is now live. Same architecture...

@MTSlive: SITUATION EXPLAINED: DeepSeek V4 Flash now beats its own Pro model, from post-training alone.

• Pri...

@Prince_Canuma: Already on MLX-VLM 🚀\n\nQT @deepseek_ai: 🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta...
@altryne: The whale resurfaces with a massive update to DeepSeek v4 Flash (beating their pro model on every be...

7. Kyrannio (Group Score: 269.4 | Individual: 44.7)

Cluster: 9 tweets | Engagement: 1298 (Avg: 205) | Type: Tech

RT @dreamina_ai: 🎬 Global Launch: Dreamina Seedance 2.5 is now live!

Our most cinematic video model yet has arrived, and Dreamina AI is your official platform. Step into a new era of precision control and long-take storytelling: · Native 30s videos in a single generation · A new interactive editing experience: reshape any part of the frame · Long video mode: up to 3 minutes, with consistency that holds throughout · Dreamina AI plugins for Maya and Blender, built for film-grade production pipelines · Up to 50 multimodal references, now with support for 3D white models and green-screen footage — bringing AI generation into real production workflows · More true-to-life lighting and shadows, richer motion detail, and precise timestamp control down to 1 second

Dreamina Seedance 2.5 is more than a model. It's an upgrade to what you can create. Now open to Dreamina AI subscribers.

🕐 Available now across Southeast Asia, the Middle East, Africa, Europe, and South America. More countries and regions coming soon. ✨ Watch @yingjian355352 in action with the new model ⏳ Limited time: save up to 32% on credits per generation 🎁 Repost and comment – first 300 users get 1,000 credits in 12 hours

#DreaminaSeedance25 #DreaminaAI

See 8 related tweets

@capcutapp: CapCut is bringing Dreamina Seedance 2.5 to creators, taking video creation to the next level. Now a...
@FellMentKE: Campaign production becomes much more powerful when AI can understand the full creative context—not ...
@alex_prompter: Stitching short clips together and hoping the character survives each cut has been the whole AI vide...
@testingcatalog: Dreamina Seedance 2.5 is now available on Dreamina AI for paid plans in Southeast Asia, the Middle E...
@AndrewCurran_: Seedance 2.5 is going rolling out now. I just checked, and it is not up for me yet. But if you live ...

8. heyshrutimishra (Group Score: 212.3 | Individual: 32.4)

Cluster: 7 tweets | Engagement: 61 (Avg: 828) | Type: Tech

H3 is the first real editor

Every other top model i know generates from a blank slate but this one i realised takes an existing clip and edits on instruction in one call.

That's a different product category tbh.

The tools doing this work right now are After Effects and DaVinci Resolve but they take technical operators & hours of iteration. H3 compresses that to a prompt, at $0.13 per second of 2K video. 👏\n\nQT @ArtificialAnlys: MiniMax H3 is #1 in Video Editing and ranks top 3 in both Text to Video and Image to Video on Artificial Analysis. MiniMax plans to release its weights under the MiniMax Community License, which would make it the leading open weights model by far

MiniMax H3 is the latest video model from MiniMax and the successor to their Hailuo family. The model has a multimodal architecture, accepting text, images, videos, and audio clips as input to generate 5-15s clips at 24fps with native audio generation. Video input support also makes MiniMax H3 one of the few models capable of instruction-based video editing, where it takes the #1 spot in our Video Editing Leaderboard. The model is also #2 in Text to Video, and #3 in Image to Video.

MiniMax H3 is available now in the Hailuo AI app and via the MiniMax API. MiniMax lists MiniMax H3 at $0.13 per second of 2K video ($ 7.80 per minute), with a 768p tier at $0.09 per second ($ 5.40 per minute) marked as coming soon in their documentation. The first five reference images are free, with each additional image at $0.04 and reference audio free. Our evaluation was conducted on the 2K tier.

At $7.80 per minute with audio, MiniMax H3 at 2K ($ 7.80/min) undercuts Dreamina Seedance 2.0 1080p ( $22.45/min), HappyHorse-1.1 ($ 9.90/min), and Kling 3.0 1080p ( $20.16/min), but is still at a premium over Google’s Gemini Omni Flash at$ 6.00/min.

MiniMax plans to release the weights under the MiniMax Community License, which permits commercial use for organizations under $20M in revenue, subject to prominent attribution requirements. If released, MiniMax H3 would become the strongest open weights video model, well ahead of the previous open weights leader LTX-2.3.

We’ll be sharing more insights into our evaluation soon - congratulations to @Hailuo_AI on the release!

See below for comparisons between MiniMax H3 and other leading models in the Artificial Analysis Video Editing Arena 🧵

See 6 related tweets

@MTSlive: SITUATION EXPLAINED: MiniMax released H3 with open weights, and it's already #1 in video editing.

•...

@heyshrutimishra: MiniMax H3 just became the best video editor in AI.

#1 on Artificial Analysis's Video Editing Leade...

@MiniMax_AI: From a single clip to finished content with sound, characters, and creative intent intact.

MiniMax ...

@testingcatalog: MiniMax H3 is now available on HailuoAI & MiniMax APIs!

All-in-One Reference - Creating Videos Us...

@alex_prompter: MiniMax H3 landed on @TopviewAIhq and the pricing is the story.

It generates native 2K, 15 second c...

9. daleverett (Group Score: 200.5 | Individual: 34.5)

Cluster: 9 tweets | Engagement: 74 (Avg: 14) | Type: Tech

Just launched Polygres on Producthunt! Turn your database into a context window for AI :D

https://t.co/Ijny5fHtaV\n\nQT @daleverett: If you’re still unsure what to film for your @ycombinator demo video, here’s ours!

If you haven't already heard, we’re @polygres, and we turn your entire database into a context window.

Yall roasted me for my heavy breathing in our founder video, fair warning i breathe heavy here too 🥺

Free for a limited time, sign up now and get free credits at full launch :D

@daleverett @daltonmeon @damienhe

See 8 related tweets

@daleverett: VC: What’s your go-to-market strategy? Me: 🥤👏🥤👏🥤\n\nQT @daleverett: I made a cringy silly Cup Song p...
@daleverett: That was fast af 😭 time to film more cringy shit ig https://t.co/dgp96iAWRN\n\nQT @daleverett: I mad...
@daleverett: Now nobody will ever forget what polygres does\n\nQT @daleverett: I made a cringy silly Cup Song par...
@polygres: Dale you are a brave man. That is cringy af.\n\nQT @daleverett: I made a cringy silly Cup Song parod...
@polygres: BROOO IM DYING FR\n\nQT @daleverett: I made a cringy silly Cup Song parody to explain our database i...

10. eng_khairallah1 (Group Score: 194.9 | Individual: 35.5)

Cluster: 8 tweets | Engagement: 105 (Avg: 55) | Type: Tech

Andrew Ng just released a free 1-hour course on building agentic knowledge Graphs from scratch:

• 00:00 - Introduction to agentic knowledge Graphs • 03:07 - Building a graph from scratch • 14:00 - Architecture of multi-agent systems • 23:00 - Building a real one with Google ADK • 01:06:03 - Why Graphsare the future of agentic AI

This 1-hour workshop will replace 10 paid courses on agentic engineering.

Watch it today, then learn how to become a Graph Engineer in the article below.\n\nQT @eng_khairallah1: https://t.co/PAT74un2ld

See 7 related tweets

@0xCodez: Head of Google AI:

"75% of our engineers are running dozens or hundreds of agents.

The way yo...

@0xMovez: Anthropic engineer:

“70% of our engineers are using self‑improving loops. Now everyone is build...

@KirkDBorne: "Building AI Agents with LLMs, RAG, and Knowledge Graphs — A practical guide to autonomous and moder...
@Av1dlive: RT @Av1dlive: Google just released free 1-hour course to master graph engineering from 0% to 100%: ...
@eng_khairallah1: RT @eng_khairallah1: Andrew Ng just released a free 1-hour course on building agentic knowledge Grap...

11. 0x_kaize (Group Score: 182.3 | Individual: 44.3)

Cluster: 7 tweets | Engagement: 172 (Avg: 46) | Type: Tech

You can get DeepSeek V4 Flash for FREE

how to get:

go to: https://t.co/5QVDDxkUyJ
sign uo
in the filters, select "Free" models
grab API key and start using

DeepSeek V4 Flash is the model that just beat V4-Pro-Preview on Agent benchmarks - Terminal Bench 82.7, Cybergym 76.7, DeepSWE 54.4

sure, it's a free tier, but even with that you can fully test the model that just forced OpenAI

don't say these FREE things are a waste of time - it's FREE after all

many people will say that requests to the original take anywhere from a few milliseconds to 1-2 seconds, while requests to these models take about 30 seconds or more. but that’s because so many people are using them

i think waiting 30 seconds to try out a new model without spending any money is the best option

as long as the model isn't too heavily loaded, it will run quickly\n\nQT @0x_kaize: this is INSANE, seriously

DeepSeek just released V4 Flash Official API and the Agent benchmarks now BEAT V4 Pro

Codex integration - fully adapted, ready to plug in
native responses API support
1M context, 284B params (13B active)

pricing that's absurd:

$0.14 /$ 0.28 per 1M tokens
$0.028 with cache hits
100x cheaper than Opus 4.8

for context, GPT-5.6 Luna JUST got cut to $0.20 /$ 1.20 - and V4 Flash is still cheaper on output

OpenAI cuts prices 80% and DeepSeek releases something 4x cheaper the next day. this is the price war escalating in real time

See 6 related tweets

@heyshrutimishra: DeepSeek just made their smaller model outperform their flagship.

V4-Flash went live today. 284B pa...

@TeksEdge: This is crazy! DeepSeek V4 Flash GA scores a 50, the same as Gemini 3.6 Flash, and one point below G...
@mervenoyann: DeepSeek V4 Flash is out and ahead of Pro and closed models per cost per task 😍 (in API)

it's a we...

@victormustar: RT @ArtificialAnlys: DeepSeek V4 Flash 0731 scores 50 on the Artificial Analysis Intelligence Index,...
@wallstengine: DEEPSEEK RELEASES V4 FLASH BETA

The Chinese AI developer’s text-only model retains a 1M-token conte...

12. Reuters (Group Score: 179.5 | Individual: 41.1)

Cluster: 7 tweets | Engagement: 578 (Avg: 80) | Type: Tech

Exclusive: Chinese military researchers have used outputs from leading US artificial intelligence models developed by OpenAI and Anthropic to train domestic AI systems to advance China's defense capabilities https://t.co/vfg5zEY9qw

See 6 related tweets

@peterwildeford: Chinese AI companies use distillation to make their products better than they otherwise would be, un...
@MarioNawfal: 🇨🇳🇺🇸 China's military may be learning from America's AI, without building it from scratch

Why spend...

@clashreport: Chinese military-linked researchers are using OpenAI and Anthropic models to train smaller domestic ...
@theobearman: Yet more evidence that adversarial distillation from Chinese presents a clear and present national s...
@Reuters: EXCLUSIVE: Chinese military researchers tap US AI models to train defence systems https://t.co/QuqJn...

13. jason (Group Score: 179.0 | Individual: 38.8)

Cluster: 7 tweets | Engagement: 1831 (Avg: 335) | Type: Tech

Open Source is winning bigly

Tokens will be 100x cheaper in 24 months

You’re going to run 50% of tokens on your local hardware unmetered

@Dell, @nvidia and @Apple are the big winners in this trend\n\nQT @deepseek_ai: 🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta!

Check out the configuration details in our official API docs: https://t.co/smCwQZMeiq

See 6 related tweets

@jason: It’s happening

Default will be local model by next year for most corporate workers\n\nQT @deepseek...

@teortaxesTex: > these scores > our upcoming DeepSeek Harness (minimal mode) Who was saying that DeepSeek doe...
@zephyr_z9: Big whale drop\n\nQT @deepseek_ai: 🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta!

🔷 W...

@AlexFinn: You can now host frontier AI on average hardware

Deepseek v4 Flash has arrived and it BEATS Fable o...

@quxiaoyin: A race to the bottom\n\nQT @deepseek_ai: 🚀 DeepSeek-V4-Flash Official API is now LIVE in public beta...

14. firstadopter (Group Score: 173.3 | Individual: 34.3)

Cluster: 6 tweets | Engagement: 96 (Avg: 158) | Type: Tech

It amazes me how disingenuous the average financial reporter is these days when they don't even mention the obvious hyperscaler capex-to-FCF cycle or the surging AI compute profit growth when putting out their sensational "oh no! freak out about current FCF" AI skeptic narratives\n\nQT @firstadopter: Since many of you seem to have the memory of a goldfish, here's how the hyperscaler capex → free cash flow cycle works as explained by Andy Jassy back in April. He's right.

Amazon: "We have customer commitments that make our capex investments predictable. We’re not investing approximately $200 billion in capex in 2026 on a hunch. The recent OpenAI commitment (over$ 100 billion) is an example of this, but there are several other customer agreements completed (and unannounced), or deep in process. Of the AWS capex we expect to spend in 2026, much of which will be monetized in 2027-2028, we already have customer commitments for a substantial portion of it.

We are willing to make large capex investments and endure short-term FCF headwinds for the substantial medium to long-term FCF surplus.

AI is a once-in-a-lifetime opportunity where the current growth is unprecedented and the future growth even bigger. AWS has a significant leadership position with the broadest functionality, strongest security and operational performance, largest share of customers and revenue, strong desire from customers to run their AI in AWS, and an opportunity to build what could be a new pillar for Amazon in chips.

We’re not going to be conservative in how we play this—we’re investing to be the meaningful leader, and our future business, operating income, and FCF will be much larger because of it."

See 5 related tweets

@StockMKTNewz: The four largest hyperscalers are expected to spend ~$725 billion on CAPEX in 2026, up ~77% from las...
@firstadopter: Narrator: Andy Jassy was right.\n\nQT @firstadopter: Since many of you seem to have the memory of a ...
@jukan05: 【Tianfeng Securities Overseas Tech】Amazon Earnings Commentary: A Virtuous Cycle Is Beginning

Amazon...

@firesidealpha: Amazon CEO Andy Jassy reveals raised 2026 capex forecasts will still not be enough to meet demand fo...
@MilkRoadAI: RT @KyleReidhead: This is the most important chart in the entire AI trade right now (Save this)

The...

15. DeItaone (Group Score: 172.7 | Individual: 33.0)

Cluster: 8 tweets | Engagement: 434 (Avg: 631) | Type: Tech

MOONSHOT BUILDS AI ON ALIBABA'S NVIDIA CLUSTER

Chinese AI startup Moonshot built its Kimi model using access to an Alibaba computing cluster of around 20,000 Nvidia GPUs, according to Bloomberg.

The partnership highlights Alibaba's role in supplying large-scale AI infrastructure as competition among Chinese AI developers intensifies.\n\nQT @DeItaone: $NVDA - MOONSHOT'S KIMI BUILT ON 20,000 NVIDIA CHIP CLUSTER VIA ALIBABA

See 7 related tweets

@WhaleInsider: JUST IN: Moonshot AI’s Kimi built a cluster of 20,000 $NVDA chips through Alibaba - Bloomberg. https...
@business: Chinese AI champion Moonshot has a computing power agreement with Alibaba for the use of around 20,0...
@coinbureau: BREAKING:🇨🇳China’s Moonshot AI is reportedly using computing power from around 20,000 Nvidia chips t...
@StockSavvyShay: Moonshot AI designed Kimi K3 (2.8T-parameter frontier model) to use less compute and move less data ...
@rohanpaul_ai: Bloomberg reports that Moonshot runs its Kimi models on roughly 20,000 Nvidia chips it accesses thro...

16. aakashgupta (Group Score: 167.5 | Individual: 33.8)

Cluster: 6 tweets | Engagement: 32 (Avg: 64) | Type: Tech

OpenAI just cut GPT-5.6 Luna's price 80%, three weeks after launch.

And the cost reduction mechanism is wild. The model reduced its own costs.

GPT-5.6 Sol rewrote OpenAI's production GPU kernels and speculative decoding inside Codex. Work that normally takes inference teams quarters to ship. The savings went straight into today's pricing.

Luna went from $1 per million input tokens and$ 6 per million output to $0.20 and$ 1.20. Terra dropped 20% to $2 and$ 12. That puts an OpenAI frontier-series model under Gemini 3.5 Flash-Lite for the first time.

If models can optimize their own serving stack, the cost curve compounds. Every generation gets a smarter optimizer working on its own infrastructure, which makes serving cheaper, which frees compute for the generation after.

Now look at what stayed expensive. Sol is unchanged at $5 and$ 30, and OpenAI added a Fast mode charging 2x the price for 2.5x the speed. Utility-tier intelligence deflated 80% in three weeks. Latency became a premium product.

Price cuts used to be a business decision. OpenAI just turned them into a model capability.\n\nQT @OpenAI: We are committed to pushing the model frontier across cost efficiency, capability, and speed.

Starting today, we are reducing prices for GPT-5.6 Luna by 80% and GPT-5.6 Terra by 20% , and offering a faster option for GPT-5.6 Sol in the API.

Luna and Terra’s lower prices are reflected in how usage is counted in Codex and ChatGPT Work, so your usage goes further.

See 5 related tweets

@Malay4Product: GPT-5.6 launched on July 9. Luna was $1 per million input tokens and$ 6 per million output. From tod...
@MTSlive: SITUATION EXPLAINED: GPT-5.6 prices dropped 80%, partly because the model optimized its own inferenc...
@cryptopunk7213: how much do you want to bet this is because openAI used a powerful unreleased model to make GPT 5.6 ...
@jerryjliu0: OpenAI’s entire growth loop: new models, price cuts, and Tibo’s token resets\n\nQT @sama: major pric...
@minchoi: Chinese AI models just got put on notice.

OpenAI slashed GPT-5.6 prices.

Luna: 80% cheaper. Terra:...

17. evanreiser (Group Score: 165.4 | Individual: 38.3)

Cluster: 5 tweets | Engagement: 46 (Avg: 31) | Type: Tech

Last week an OpenAI model escaped its sandbox and hacked Hugging Face. Today Anthropic just disclosed three incidents where Claude models reached the open internet and hacked into three real companies.

This shouldn't shock anyone who has vibe-coded. You've watched an AI confidently delete the wrong file/database. AI makes mistakes.

Give that same intelligence superhuman persistence, machine speed, and a pathway to production systems, and its no longer a tiny coding bug, it starts causing real damage in the world.

One example from the disclosure: believing it was still in a simulation, Claude attacked its target by publishing a malware into a real public code registry. That required an account. The account required an email. The email required a phone number. It tried to find money to buy one, failed, backtracked to a free email provider, registered, and uploaded the malware.

None of the victims noticed. Presumably they learned of the breaches when Anthropic called them, in some cases months later. Obviously: credit to Anthropic and OpenAI for auditing themselves and publishing the details so we can learn as a global cyber community.

Unfortunately, what just happened is the best case: an aligned model with safety training, pursuing a harmless test goal, no criminal involved. What happens when criminals start steering frontier models on purpose? Or we have purpose-built malicious models with no safeguards?

Every one of these attacks WAS detectable, though. BUT it took some someone noticing something that was weird, strange or Abnormal:

Hugging Face noticed anomalous activity. OpenAI's monitoring flagged unusual behavior. Anthropic found its incidents by reviewing transcripts for unexpected actions.

None came from recognizing a known attack or something that has been seen before. All came from noticing behavior that wasn't normal.

Attackers, human or AI, have one unavoidable weakness: they have to look normal every time. They can't. But no human reviews logs at machine speed (or machine efficiency).

Exploiting that weakness requires AI-native defense: baseline normal behavior, detect anomalies, respond autonomously in real time.

We built @Abnormal Behavioral AI to detect these type of cyber attacks and response at machine speed. Now the attacks are arriving faster than we can track, and certainly faster than any security roadmap assumed.

Now is the time to rally across the cyber community to ensure that we're using good AI to stop the bad AI before it gets totally out of control.\n\nQT @AnthropicAI: In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.

Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews.

See 4 related tweets

@BrianRoemmele: “Our AI did a bad thing, we are so worried”—Anthropic

Solution:

They would rather drain the ocean...

@BrianRoemmele: ANTHROPIC AGAIN—WHEN WILL THEY LEARN?  Training Frontier Models on Internet Sewage Keeps Producing S...
@MTSlive: SITUATION EXPLAINED: Anthropic found three incidents where Claude reached real production systems fr...
@Abnormal: RT @evanreiser: Last week an OpenAI model escaped its sandbox and hacked Hugging Face. Today Anthrop...

18. StockMKTNewz (Group Score: 162.4 | Individual: 34.5)

Cluster: 6 tweets | Engagement: 178 (Avg: 283) | Type: Tech

JPMorgan today raised its price target on Amazon $AMZN up to$ 365 from $330 while maintaining its Overweight rating

Wells Fargo raised its price target on Amazon $AMZN up to$ 328 from $322 while maintaining its Buy rating

UBS raised its price target on Amazon $AMZN up to$ 318 from $305 while maintaining its Buy rating

Biard raised its price target on Amazon $AMZN up to$ 310 from $300 while maintaining its Outperform rating

Citi raised its price target on Amazon $AMZN up to$ 350 from $325 while maintaining its Buy rating

Barclays raised its price target on Amazon $AMZN up to$ 365 from $330 while maintaining its Overweight rating

Pivotol Research raised its price target on Amazon $AMZN up to$ 333 from $320 while maintaining its Buy rating

TD Cowen raised its price target on Amazon $AMZN up to$ 350 from $340 while maintaining its Buy rating

Cantor Fitzgerald lowered its price target on Amazon $AMZN down to$ 320 from $330 while maintaining its Overweight rating\n\nQT @TipRanks: Analysts were broadly bullish on Amazon$ AMZN following its Q2 results, with most firms raising their price targets and reiterating Buy or Overweight ratings.

The main takeaway was AWS’s accelerating growth, driven by expanding enterprise AI adoption, rising cloud capacity, and a rapidly growing backlog. Analysts highlighted Amazon’s $25 billion in AI annual recurring revenue, its$ 496 billion backlog, stronger-than-expected AWS margins, and clear operating leverage as evidence that heavy AI investment is beginning to generate meaningful returns.

JPMorgan and Barclays raised their targets to a Street-high $365, while Citi and TD Cowen moved to$ 350. Wells Fargo reaffirmed Amazon as a top pick, expecting capacity expansion to support further growth in 2027. Pivotal also pointed to “blowout” cloud and advertising results.

Cantor Fitzgerald was the only firm to lower its target – to $320 from$ 330 – though it maintained an Overweight rating and remained positive on AWS’s long-term momentum. Overall, analysts viewed Amazon’s increased capex as justified by strong AI demand, improving margins, and compelling expected returns.

See 5 related tweets

@wallstengine: Bernstein Raises $AMZN PT to$ 320 from $315 - Outperform

Analyst comments: If you liked MSFT’s prin...

@StockSavvyShay: $AMZN posts its biggest one-day gain in 14 years 🙃 https://t.co/QLyCppBNhF\n\nQT @StockSavvyShay:$ ...
@wallstengine: Morgan Stanley on $AMZN (OW, PT$ 335)

Analyst comments: "AWS's approximately 840bps growth accelera...

@DeItaone: $AMZN - BERNSTEIN RAISES AMAZON PT ON AWS, AI MOMENTUM

Bernstein raised its Amazon price target to...

@BullTheoryio: BREAKING: $AMZN is up 9% today after AWS cloud revenue grew 37% to$ 42.2 BILLION, its fastest growth...

19. FirstSquawk (Group Score: 162.2 | Individual: 26.6)

Cluster: 10 tweets | Engagement: 207 (Avg: 36) | Type: Tech

ANTHROPIC HAS REVEALED THAT ITS AI MODELS ESCAPED TESTING ENVIRONMENTS AND HACKED THREE UNSUSPECTING COMPANIES WHEN INADVERTENTLY GRANTED INTERNET ACCESS, GOING UNDETECTED SINCE APRIL

See 9 related tweets

@TechCrunch: Anthropic says its own AI models breached three companies during security tests https://t.co/m0ASEjE...
@Reuters: Anthropic's AI hacked three companies during tests, highlighting growing security risks https://t.co...
@unusual_whales: BREAKING: Anthropic announced that its artificial intelligence models had breached three different o...
@Forbes: Anthropic Says Its AI Models Hacked Into Three Organizations During Testing https://t.co/nQWzBqwff5 ...
@clashreport: Anthropic Reveals Rogue AI Models Hacked Three Companies During Testing

Anthropic disclosed that it...

20. XFreeze (Group Score: 162.2 | Individual: 35.8)

Cluster: 5 tweets | Engagement: 174 (Avg: 393) | Type: Tech

Europe’s AI strategy is finally working

Russia might actually have competition now 😂\n\nQT @vonderleyen: AI is the most important technology of our time.

Europe wants to become the first AI Continent.

For advanced healthcare, for the transport sector and so much more.

European AI Gigafactories will provide the necessary computing power to make this possible.

Together with our Member States, we are funding their construction with up to €10 billion, which are set to unlock at least €20 billion in private investements across the EU.

Together, we are building our technological sovereignty.

https://t.co/Vxko4TRXVL

See 4 related tweets

@scaling01: > "Europe wants to become the first AI Continent" > EU announces €10B for new datacenters

mfw...

@michalmalewicz: €1B will go towards buying a couple of GPUS

Then another €1B for import fees, the EU customs will h...

@ianmiles: Spain needs more migrants to fill the roles of it's AI developers since most of them already left.\n...
@ianmiles: Europe has achieved the milestone of being ths first ever continent to create a comprehensively set ...