热门技术推文 - 2026年4月10日

2026年4月10日科技每日简报

Today's top tech conversations are led by @matteocollina, whose post about 'RT @OpenAI: We’re updating our...' garnered the highest engagement. Key themes trending across the top stories include https, codex, model, models, openai. The community is actively discussing recent developments in AI, engineering practices, and startup strategies.

1. matteocollina (Group Score: 447.1 | Individual: 56.5)

Cluster: 15 tweets | Engagement: 1560 (Avg: 145) | Type: Tech

RT @OpenAI: We’re updating our ChatGPT Pro and Plus subscriptions to better support the growing use of Codex.

We’re introducing a new $100/month Pro tier. This new tier offers 5x more Codex usage than Plus and is best for longer, high-effort Codex sessions.

In ChatGPT, this new Pro tier still offers access to all Pro features, including the exclusive Pro model and unlimited access to Instant and Thinking models.

To celebrate the launch, we’re increasing Codex usage for a limited time through May 31st so that Pro $100 subscribers get up to 10x usage of ChatGPT Plus on Codex to build your most ambitious ideas.

See 14 related tweets

@kimmonismus: $100 ChatGPT pro tier official.

-5x codex rates -access to ChatGPT pro -unlimited thinking

Yeah,...

@burkov: OMG! SWITCHING FROM CLAUDE CODE TO CODEX NOW!\n\nQT @OpenAI: We’re updating our ChatGPT Pro and Plus...
@OpenAI: Our existing $200 Pro tier still remains our highest usage option. And as a thank you to our existin...
@badlogicgames: just need to ask nicely, really.\n\nQT @OpenAI: We’re updating our ChatGPT Pro and Plus subscription...
@testingcatalog: OpenAI introduced a new $100/month Pro plan with 5x more Codex usage. A new Pro tier also includes a...

2. edzitron (Group Score: 293.6 | Individual: 40.9)

Cluster: 11 tweets | Engagement: 2667 (Avg: 659) | Type: Tech

https://t.co/lhrlbP2o7W\n\nQT @wallstengine: Axios: OpenAI is planning a staggered rollout for a new model with advanced cybersecurity capabilities, limiting access to a small group of companies over fears the tool could be misused. The move would mirror Anthropic’s restricted release of Mythos. https://t.co/vhENYDWBSb

See 10 related tweets

@zephyr_z9: BRUH... The gates are closing down\n\nQT @wallstengine: Axios: OpenAI is planning a staggered rollou...
@wallstengine: Axios: OpenAI is planning a staggered rollout for a new model with advanced cybersecurity capabiliti...
@jukan05: As models become more advanced, it will become more likely that their release to retail users is del...
@victormustar: There's a high chance you'll NOT get access to the most intelligent models.

when we say open source...

@Techmeme: Source: OpenAI is finalizing a model with advanced cybersecurity capabilities that it plans to relea...

3. rickasaurus (Group Score: 236.2 | Individual: 34.9)

Cluster: 10 tweets | Engagement: 2330 (Avg: 363) | Type: Tech

RT @claudeai: We're bringing the advisor strategy to the Claude Platform.

Pair Opus as an advisor with Sonnet or Haiku as an executor, and get near Opus-level intelligence in your agents at a fraction of the cost. https://t.co/fRkegyMs5t

See 9 related tweets

@aakashgupta: This diagram is the entire AI agent cost problem solved in one architecture.

Every company building...

@cryptopunk7213: this is a banger, anthropic just gave everyone a tool that makes their cheapest model perform like t...
@scaling01: advisor tool benchmarks

idea: Opus plans, smaller models execute

it's cheaper and better than usin...

@arvidkahl: The delegate starts delegating.

This is what I love about AI orchestration, and I thinks it’s only...

@dejavucoder: >wake up >dread over mythos >use claude for work >claude drops new feature >sleep\n\n...

4. badlogicgames (Group Score: 196.4 | Individual: 49.9)

Cluster: 6 tweets | Engagement: 1082 (Avg: 117) | Type: Tech

RT @karpathy: Judging by my tl there is a growing gap in understanding of AI capability.

The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. This is a group of reactions laughing at various quirks of the models, hallucinations, etc. Yes I also saw the viral videos of OpenAI's Advanced Voice mode fumbling simple queries like "should I drive or walk to the carwash". The thing is that these free and old/deprecated models don't reflect the capability in the latest round of state of the art agentic models of this year, especially OpenAI Codex and Claude Code.

But that brings me to the second issue. Even if people paid $200/month to use the state of the art models, a lot of the capabilities are relatively "peaky" in highly technical areas. Typical queries around search, writing, advice, etc. are not the domain that has made the most noticeable and dramatic strides in capability. Partly, this is due to the technical details of reinforcement learning and its use of verifiable rewards. But partly, it's also because these use cases are not sufficiently prioritized by the companies in their hillclimbing because they don't lead to as much $$$ value. The goldmines are elsewhere, and the focus comes along.

So that brings me to the second group of people, who both 1) pay for and use the state of the art frontier agentic models (OpenAI Codex / Claude Code) and 2) do so professionally in technical domains like programming, math and research. This group of people is subject to the highest amount of "AI Psychosis" because the recent improvements in these domains as of this year have been nothing short of staggering. When you hand a computer terminal to one of these models, you can now watch them melt programming problems that you'd normally expect to take days/weeks of work. It's this second group of people that assigns a much greater gravity to the capabilities, their slope, and various cyber-related repercussions.

TLDR the people in these two groups are speaking past each other. It really is simultaneously the case that OpenAI's free and I think slightly orphaned (?) "Advanced Voice Mode" will fumble the dumbest questions in your Instagram's reels and at the same time, OpenAI's highest-tier and paid Codex model will go off for 1 hour to coherently restructure an entire code base, or find and exploit vulnerabilities in computer systems. This part really works and has made dramatic strides because 2 properties: 1) these domains offer explicit reward functions that are verifiable meaning they are easily amenable to reinforcement learning training (e.g. unit tests passed yes or no, in contrast to writing, which is much harder to explicitly judge), but also 2) they are a lot more valuable in b2b settings, meaning that the biggest fraction of the team is focused on improving them. So here we are.

See 5 related tweets

@rohanpaul_ai: AI’s most confusing feature right now is that the strongest systems are genuinely terrifying in nar...
@Scobleizer: After building with bleeding edge AI I get this separation that @karpathy lays out deeply.

Family a...

@aakashgupta: Karpathy just gave you the most concise explanation of why AI feels like two completely different te...
@garrytan: You need to use frontier models with giant context and actually have systems that give them the righ...
@tunguz: Exactly right. If you are using AI for anything technical, you are flabbergasted by the advancement ...

5. wallstengine (Group Score: 183.8 | Individual: 46.9)

Cluster: 5 tweets | Engagement: 408 (Avg: 93) | Type: Tech

$AMZN CEO Andy Jassy issues annual shareholder letter.

Here are 8 things to know:

Committing over $4B to expand its rural delivery network
Now has 1M+ robots operating in its fulfillment network
AWS AI revenue is now running at $15B+ annually
Amazon’s chips business, including Graviton, Trainium, and Nitro, is now at a $20B+ annual run rate
Expects about $200B of capex in 2026, largely tied to AI Infra
Trainium2 is largely sold out, Trainium3 is nearly fully subscribed, and part of Trainium4 is already reserved
Amazon said its grocery business reached $150B+ in gross sales in 2025
Amazon Leo already has 200+ satellites in orbit ahead of its mid-2026 launch

See 4 related tweets

@StockSavvyShay: $AMZN is up ~5% after CEO Andy Jassy said two large AWS customers wanted all of Amazon’s 2026 Gravit...
@StockSavvyShay: $AMZN CEO Andy Jassy released Amazon’s annual shareholder letter:

• AWS AI revenue is running above...

@StockSavvyShay: The wildest part is that Jassy says $AMZN chip business would be worth ~$ 50B if sold externally inst...
@Techmeme: Annual letter: Andy Jassy says AWS's AI revenue has reached a $15B annual run rate as of Q1, and Ama...

6. chhddavid (Group Score: 146.6 | Individual: 46.0)

Cluster: 4 tweets | Engagement: 189 (Avg: 23) | Type: Tech

this is just f*king scary......\n\nQT @chddaniel: Introducing Shipper. The only Claude-first AI Business Builder.

Claude Opus 4.6 researches your competitors, then builds your app: code, design, monetization, launch, translation, emails etc.

It's also the only Opus-class app builder. https://t.co/kLj3IJaYJO

See 3 related tweets

@chhddavid: so you're telling me Claude Code Opus 4.6 can now...
deep research my competitors
build me a fu...
@Shipper_now: this is actually horrifying...\n\nQT @chddaniel: Introducing Shipper. The only Claude-first AI Busin...
@chhddavid: so you're telling me Claude Opus 4.6 can now...
scan competition
oneshot their $1B app
launch...

7. paoloardoino (Group Score: 144.1 | Individual: 33.6)

Cluster: 7 tweets | Engagement: 304 (Avg: 565) | Type: Tech

The world is approaching a moment where billions of humans share the planet with billions of autonomous machines and trillions of AI agents.

The current model, routing every decision through a centralized server, won’t scale to meet that reality. The laws of physics alone make centralized AI a dead end: speed-of-light latency, single points of failure, and concentration of control are features of a system designed for a smaller world.

QVAC is built for the world that’s coming. QVAC is the fundamental building block in the era of Stable Intelligence.\n\nQT @tether: Tether Launches QVAC SDK as the AI Universal Building Block that Runs, Trains, and Evolves Intelligence Across any Device and Platform Learn more: https://t.co/n7mkjkD8cz

See 6 related tweets

@paoloardoino: At Tether, we see AI as a new element of the periodic table – a raw material that can be embedded i...
@paoloardoino: RT @qvac: The engine of the 21st century is here. 🧠

The QVAC SDK is the "steam engine" of the AI er...

@Cointelegraph: 🔥 LATEST: Tether launches open-source QVAC SDK for cross-platform AI development and deployment. htt...
@Cointelegraph: ⚡️ NEW: Tether has launched its open-source QVAC SDK, a cross-platform AI framework that lets develo...
@BitcoinNews: 🗞️ Tether launches QVAC SDK, a local-first, modular AI solution.

The @qvac SDK enables decentralize...

8. CoreWeave (Group Score: 140.8 | Individual: 28.6)

Cluster: 7 tweets | Engagement: 85 (Avg: 32) | Type: Tech

CoreWeave and @Meta expand their collaboration with a $21B agreement through 2032, building on$ 14.2B from September — bringing the total announced with Meta to ~$35B.

A clear signal: demand is accelerating, driven by inference. https://t.co/yJA9r4JPuu

See 6 related tweets

@wallstengine: $META has committed an additional$ 21B to CoreWeave $CRWV for AI cloud infrastructure, adding to a p...
@business: CoreWeave has expanded its deal to supply artificial intelligence computing power to Meta to $21 bil...
@wallstengine: $META COMMITS TO SPENDING ADDITIONAL$ 21 BILLION WITH COREWEAVE $CRWV AS AI COSTS KEEP RISING https:...
@Techmeme: Meta commits to spending additional $21B on AI cloud infrastructure from CoreWeave, running from 202...
@CNBC: Meta commits to spending additional $21 billion with CoreWeave as AI costs keep rising https://t.co/...

9. alex_prompter (Group Score: 140.2 | Individual: 34.5)

Cluster: 6 tweets | Engagement: 7 (Avg: 278) | Type: Tech

Meta spent $115-135 BILLION on AI infrastructure last year.

Hired Scale AI’s CEO. Built a whole new lab. Abandoned open source.

The result? “Muse Spark” - a model that’s “as capable as their older midsize Llama 4.”

They burned the GDP of a small country to match their own mid-tier model from last year.

And they’re calling it a win.\n\nQT @AIatMeta: Introducing Muse Spark, the first in the Muse family of models developed by Meta Superintelligence Labs.

Muse Spark is a natively multimodal reasoning model with support for tool-use, visual chain of thought, and multi-agent orchestration.

Muse Spark is available today at https://t.co/wHkMPH82ZH and the Meta AI app. We’re also making it available in private preview via API to select partners, and we hope to open-source future versions of the model.

Learn more: https://t.co/PloE9q5x96

See 5 related tweets

@WesRoth: Muse Spark hit a 52 on the Artificial Analysis Intelligence Index, placing it firmly in the top 5 mo...
@QuixiAI: I asked it to help me architect an open source alternative to Mythos and it refused, claiming that M...
@eladgil: 👀\n\nQT @alexandr_wang: 1/ today we're releasing muse spark, the first model from MSL. nine months a...
@adonis_singh: muse-spark is a really interesting model, such a shame that we can't use it via api

meta(.)ai is a ...

@ylecun: RT @alexandr_wang: 1/ today we're releasing muse spark, the first model from MSL. nine months ago we...

10. intel (Group Score: 134.3 | Individual: 33.2)

Cluster: 6 tweets | Engagement: 442 (Avg: 1171) | Type: Tech

AI doesn't run on accelerators alone — it runs on systems 🖥️ We're deepening our collaboration with @Google to advance AI infrastructure built for the real world: Intel Xeon CPUs powering Google Cloud + expanded co-development of custom IPUs for smarter, more efficient hyperscale AI.CPUs. IPUs. Systems that scale.\n\nQT @intelnews: Intel and @Google announce a multiyear collaboration to advance AI and cloud infrastructure. 🔹 Intel® Xeon® processors continue powering Google Cloud AI, inference, and general-purpose workloads 🔹 Expanded co-development of custom ASIC-based IPUs 🔹 A balanced, heterogeneous approach to AI system design at scale

"Scaling AI requires more than accelerators — it requires balanced systems." — @LipBuTan1 CEO, Intel

11. shiri_shh (Group Score: 123.9 | Individual: 33.9)

Cluster: 5 tweets | Engagement: 98 (Avg: 201) | Type: Tech

OpenAI is panicking. The $100 plan proves it.

This week, Anthropic hit $30B ARR and passed OpenAI.

Tripled revenue in four months. Claude Code alone is doing $2.5B. 1,000+ companies paying a million a year each.

OpenAI was Stuck at $25B. Bleeding devs to Claude every single day.

So they did the only thing left.

Launched a $100 Pro Lite plan. Same price as Claude Max. Same pitch.

And to keep $200 Pro users from walking, they extended the 2x Codex promo till May 31 and reset Codex rate limits. Again.

They already killed Sora. Ended the Disney deal. Cut every side quest. Now they're mimicking Anthropic's pricing and throwing freebies at loyal users to stop the bleed.\n\nQT @OpenAI: We’re updating our ChatGPT Pro and Plus subscriptions to better support the growing use of Codex.

We’re introducing a new $100/month Pro tier. This new tier offers 5x more Codex usage than Plus and is best for longer, high-effort Codex sessions.

In ChatGPT, this new Pro tier still offers access to all Pro features, including the exclusive Pro model and unlimited access to Instant and Thinking models.

See 4 related tweets

@aakashgupta: OpenAI just cut their most expensive consumer product in half and framed it as a Codex feature.

The...

@rohanpaul_ai: OpenAI just rolled out a $100/month ChatGPT Pro plan.

So OpenAI now has 2 ChatGPT Pro plans for ind...

@gdgtify: RT @rohanpaul_ai: OpenAI just rolled out a $100/month ChatGPT Pro plan.

So OpenAI now has 2 ChatGPT...

@shiri_shh: RT @shiri_shh: OpenAI is panicking. The $100 plan proves it.

This week, Anthropic hit $30B ARR and ...

12. shiri_shh (Group Score: 121.3 | Individual: 39.6)

Cluster: 4 tweets | Engagement: 356 (Avg: 201) | Type: Tech

sir...Claude has shipped 75+ updates in 60 days and ours is just 5 sir. https://t.co/L2McHt8aN8\n\nQT @claudeai: Introducing Claude Managed Agents: everything you need to build and deploy agents at scale.

It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days.

Now in public beta on the Claude Platform. https://t.co/vHYfiC1G56

See 3 related tweets

@WesRoth: Anthropic unveiled "Claude Managed Agents," a newly released public beta platform that let developer...
@ferologics: there goes our agent infra lol\n\nQT @claudeai: Introducing Claude Managed Agents: everything you ne...
@minchoi: RT @minchoi: Anthropic just launched Claude Managed Agents.

Now anyone can build and deploy product...

13. svpino (Group Score: 121.3 | Individual: 32.0)

Cluster: 4 tweets | Engagement: 111 (Avg: 140) | Type: Tech

Oh, this is pretty interesting:

These guys built a completely new architecture: Large Memory Models.

This is designed specifically for how human memory works. Instead of RAG or vector search, this is a different paradigm.

Their founders have 160+ publications in Nature and ICLR, and closed their Harvard lab to build this.\n\nQT @gkreiman: Imagine a future where you can REMEMBER EVERYTHING. Every email, every person, every conversation. Introducing Engramme. Our vision is to endow humans with perfect and infinite memory. All your memories come to you. No more searching or prompting. https://t.co/LX22KMG2lH 🧠 https://t.co/4o7pvMfkGO

See 3 related tweets

@rohanpaul_ai: AI solved language.

It solved vision.

It solved audio.

But memory is still broken.

That’s the ga...

@Scobleizer: This is a big deal: A much more human-like memory system.

Not built on transformers. Engramme crea...

@chatgpt21: This startup is trying to be the (good) big brother so you can have infinite memory of your life htt...

14. unusual_whales (Group Score: 108.0 | Individual: 27.2)

Cluster: 4 tweets | Engagement: 158 (Avg: 2589) | Type: Tech

Good morning to everyone:\n\nQT @unusual_whales: BREAKING: Look at this.

This user built an options strike calculator using the UW API.

What do you want to build?

Link: https://t.co/kBsFwq7F8x https://t.co/UrQPUwQOOs

See 3 related tweets

@unusual_whales: Good night to everyone:\n\nQT @unusual_whales: BREAKING: Look at this.

This user built an options...

@unusual_whales: BREAKING: Look at this.

This user built an options strike calculator using the UW API.

What do...

@unusual_whales: Look at this.

This user built an options strike calculator using the UW API.

What do you want to...

15. chhddavid (Group Score: 103.0 | Individual: 37.1)

Cluster: 3 tweets | Engagement: 64 (Avg: 23) | Type: Tech

this is f*king scary guys...\n\nQT @shipper_now: Introducing Shipper

AI that one-shots any $1B company in 183 seconds.

Shipper is an army of AI agents replacing any founder skill. Build full business from 1 sentence. Claude will run it forever for you.

See 2 related tweets

@chddaniel: so you're saying Claude Code Opus 4.6 can now...
research companies worth $1B
oneshot their ent...
@chddaniel: this is actually terrifying......\n\nQT @shipper_now: Introducing Shipper

AI that one-shots any $1B...

16. aakashgupta (Group Score: 101.7 | Individual: 35.3)

Cluster: 3 tweets | Engagement: 17 (Avg: 154) | Type: Tech

This is one of those features that looks like a quality-of-life improvement and is actually an architecture change.

Before Monitor, Claude Code's agent loop worked like every other coding agent: spin, check, spin, check, burn tokens on empty polls. The agent had to stay awake to know if something happened. That's the equivalent of paying a developer to stare at a log file 24 hours a day.

Monitor flips the model. Claude sets a trigger condition, goes to sleep, and the OS wakes it up when the condition fires. tail -f | grep ERROR costs zero tokens until an error appears. The agent only burns compute when there's actual work to do.

This is the same architectural shift that turned web development from CGI polling into WebSockets. From pull to push. The agent that sleeps until poked will always beat the agent that spins in a loop, because compute costs scale with vigilance under polling and scale with incidents under Monitor.

The teams shipping Claude Code agents into production CI/CD pipelines just got their costs cut by an order of magnitude on any monitoring workflow. And Anthropic just made the persistent background agent viable at a price point where you'd actually leave it running.\n\nQT @noahzweben: Thrilled to announce the Monitor tool which lets Claude create background scripts that wake the agent up when needed.

Big token saver and great way to move away from polling in the agent loop

Claude can now:

Follow logs for errors
Poll PRs via script
and more! https://t.co/eflixzi0xk

See 2 related tweets

@noahzweben: Thrilled to announce the Monitor tool which lets Claude create background scripts that wake the agen...
@trq212: you'll need to explicitly prompt Claude Code to use it, but the Monitor Tool is super powerful

e.g....

17. wallstengine (Group Score: 100.0 | Individual: 33.2)

Cluster: 4 tweets | Engagement: 121 (Avg: 93) | Type: Tech

OpenAI Advertising Revenue, per The Information:

2026: $2.4B 2027:$ 11B $2030:$ 102B

Its ad pilot also topped $100M in annualized revenue just six weeks after launch. https://t.co/wQnoSWQLGm

See 3 related tweets

@Techmeme: Documents: OpenAI expects ads to generate ~$2.4B in 2026 revenue and to quadruple in 2027 to nearly ...
@wallstengine: OpenAI expects ad revenue to reach $2.5B this year and$ 100B by 2030. An Axios report says OpenAI’s ...
@FirstSquawk: OPENAI EXPECTS TO REACH $100 BILLION IN ADVERTISING REVENUE BY 2030, ACCORDING TO AXIOS....

18. rseroter (Group Score: 97.8 | Individual: 40.3)

Cluster: 4 tweets | Engagement: 1002 (Avg: 108) | Type: Tech

RT @GeminiApp: Gemini can now transform your questions and complex concepts into customizable interactive visualizations directly in your chat.

Adjust variables, rotate 3D models, and explore data for a more immersive way to learn and explore in Gemini.

See 3 related tweets

@IamEmily2050: Gemini team keep shipping and we still few weeks away from Google I/O.\n\nQT @GeminiApp: Gemini can ...
@testingcatalog: Gemini can now help visualize complex topics through interactive experiences directly in chat.

"Sh...

@joshwoodward: Expect even more creative model replies (like this) coming soon!\n\nQT @GeminiApp: Gemini can now tr...

19. chddaniel (Group Score: 92.7 | Individual: 31.5)

Cluster: 3 tweets | Engagement: 0 (Avg: 26) | Type: Tech

so you're telling me Claude Opus 4.6 can now...

research printing apps
clone and launch them for me
do email marketing
self-build new features

without any human interaction!??

it's so over... https://t.co/lQE3Dg5XrL https://t.co/5vhR7cgKzP\n\nQT @chhddavid: 🚨BREAKING: Someone built a money printer… you send a prompt and it just starts bringing in customers in your sleep.

It’s called Shipper.

It reads prompt, figures out who should be buying, finds similar companies, and researches what they did to print money.

I tried it on a @calai_app and it immediately pulled in strats I wouldn’t have found myself.

Here’s what happens:

→ A full web or mobile app based on your idea → The core features mapped and built automatically → Design, frontend, backend all handled → It prepares monetization (payments, flows, etc) → It can even get it ready for app stores

No dev. No setup. No figuring things out.

Here’s the wildest part:

Most people still treat building like manual work.

Planning. Designing. Coding.

This skips all of it.

You start with a sentence.

And it brings you customers.

See 2 related tweets

@Shipper_now: this is outright scary......\n\nQT @chhddavid: 🚨BREAKING: Someone built a money printer… you send a ...
@chddaniel: this is outright terrifying...\n\nQT @chhddavid: 🚨BREAKING: Someone built a money printer… you send ...

20. SIGKITTEN (Group Score: 91.1 | Individual: 46.2)

Cluster: 4 tweets | Engagement: 540 (Avg: 56) | Type: Tech

RT @ClementDelangue: "But here is what we found when we tested: We took the specific vulnerabilities Anthropic showcases in their announcement, isolated the relevant code, and ran them through small, cheap, open-weights models. Those models recovered much of the same analysis. Eight out of eight models detected Mythos's flagship FreeBSD exploit, including one with only 3.6 billion active parameters costing $0.11 per million tokens. A 5.1B-active open model recovered the core chain of the 27-year-old OpenBSD bug." https://t.co/yBTiiMq1Xy

See 3 related tweets

@WesRoth: Anthropic detailed the extreme autonomous capabilities of the unreleased "Claude Mythos Preview."

...

@teortaxesTex: RT @paul_cal: >8 out of 8 [cheap oss] models detected Mythos's flagship FreeBSD exploit

Complete...

@ylecun: RT @stanislavfort: New post: We tested the Mythos showcase vulnerabilities with open models.

They ...