GPT-5.5 release 🚀, Anthropic $1T valuation 💰, DeepSeek v4

OpenAI released GPT-5.5 with improved agentic reasoning, tool use, and
efficiency, matching prior latency while increasing performance across
coding ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

 Sign Up [1] |Advertise [2]|View Online [3]

		TLDR

		TOGETHER WITH [Crusoe] [4]

TLDR AI 2026-04-24

 WAS YOUR CLOUD BUILT FOR AI? GET UP TO 10X FASTER TTFT AND 5X BETTER
THROUGHPUT WITH CRUSOE CLOUD (SPONSOR) [4]

 AI workloads don't simply need scale, they need ultra-low latency at
high throughput. Crusoe [4]'s distributed memory architecture behaves
like it's local, even at peak demand. The result? 

👉 9.9x faster time-to-first-token and 5x better throughput than
vLLM. 

✅ Build, train, and serve AI [4] all in one place

✅ Use the model(s) of your choice

✅ Spend less time dealing with infra and more time building products

Explore Crusoe Managed Inference [4]

🚀

HEADLINES & LAUNCHES

 GPT 5.5 (18 MINUTE READ) [5]

 OpenAI released GPT-5.5 with improved agentic reasoning, tool use,
and efficiency, matching prior latency while increasing performance
across coding and knowledge tasks.

 DEEPSEEK UNVEILS FLAGSHIP AI MODEL A YEAR AFTER BREAKTHROUGH (7
MINUTE READ) [6]

 DeepSeek has rolled out new preview versions of its new flagship AI
model. The V4 Flash and V4 Pro series comes with architecture upgrades
and optimization improvements. It has a 1-million-token context
window. While the service capacity for the V4 Pro series is extremely
limited due to a computing crunch, DeepSeek expects pricing for the
model to drop once Huawei releases its Ascend 950-powered computing
clusters.

 TENCENT, ALIBABA TO BACK DEEPSEEK AT $20B+ VALUATION (2 MINUTE READ)
[7]

 DeepSeek is in talks for its first funding round at a $20 billion
valuation, with Tencent and Alibaba interested. Tencent is seeking a
20% stake, but DeepSeek doesn't want to lose that much control. The
valuation surged from $10 billion to $20 billion in days, illustrating
significant investor interest.

 ANTHROPIC JUST OVERTOOK OPENAI WITH $1 TRILLION VALUATION (2 MINUTE
READ) [8]

 Anthropic reached a $1 trillion valuation on Forge Global, surpassing
OpenAI's $880 billion. The spike in Anthropic's value stems from
scarce available shares and growing demand due to partnerships and
increased use of its Claude Code tool. Current investor interest
highlights Anthropic's position as a leading AI opportunity.

🧠

DEEP DIVES & ANALYSIS

 TRAINING FOR ACCURACY IN SEARCH LLMS (5 MINUTE READ) [9]

 Perplexity detailed a staged post-training approach that teaches
models tool use, evidence gathering, and structured evaluation to
improve accuracy in search-based AI systems.

 AGENTICS: AI ENABLEMENT REQUIRES MANAGED AGENT RUNTIMES (10 MINUTE
READ) [10]

 Enterprises need managed agent runtimes for effective AI integration,
avoiding the complexities of manual configuration. Current solutions
like Claude Code are unintuitive, leading to inefficient setups and
potential security risks. Companies like Ramp and Stripe have
developed custom solutions, but managed services are scarce, creating
opportunities for new entrants in the market.

🧑‍💻

ENGINEERING & RESEARCH

 THE SEARCH ENGINE BEHIND CURSOR, NOTION, AND LINEAR (SPONSOR) [11]

 turbopuffer is the search engine that Cursor, Notion, and Linear use
to retrieve every byte of context without blowing out margins. It's
built natively on object storage with smart caching, so it's as fast
(p90 < 20ms) as in-memory search, but way cheaper.

Read how Cursor cut costs 20x and improved agent reasoning [12]

 AN UPDATE ON RECENT CLAUDE CODE QUALITY REPORTS (11 MINUTE READ) [13]

 Anthropic listened to feedback that Claude's responses had worsened,
and it has resolved the issue as of April 20. The startup traced the
problem to three separate changes that affected Claude Code, the
Claude Agent SDK, and Claude Cowork. The API was not impacted. This
post looks at how Anthropic investigated and fixed the problem and the
commitments the company has made to ensure similar issues are much
less likely to happen again.

 OPENAI PRIVACY FILTER MODEL (8 MINUTE READ) [14]

 OpenAI released a lightweight open-weight model for detecting and
redacting PII in text, designed for fast, local, context-aware privacy
filtering workflows.

 EXPERT UPCYCLING (GITHUB REPO) [15]

 Amazon researchers released a method to expand Mixture-of-Experts
models mid-training by duplicating and specializing experts, improving
performance without increasing inference cost.

🎁

MISCELLANEOUS

 AI CODING FIRM COGNITION IN FUNDING TALKS AT $25 BILLION VALUE (3
MINUTE READ) [16]

 Cognition AI is in early talks to raise a round of funding that would
more than double its valuation to $25 billion. The talks are ongoing
and the terms could change. Cognition uses AI to streamline the
process of writing and debugging code, with a focus on selling to
businesses. Its flagship product, Devin, is being used by companies
like Anduril and Microsoft.

 ORACLE'S DELUGE OF AI DEBT PUSHES WALL STREET TO THE LIMIT (5 MINUTE
READ) [17]

 Oracle's $300 billion megadeal with OpenAI is testing the limit of
Wall Street's appetite for debt tied to the datacenter boom. Banks
have struggled for months to spread the risk of the billions of
dollars in loans they made to build data centers leased to Oracle in
Texas and Wisconsin. Bank balance sheets are now clogged, constraining
the financing prospects of future projects tied to Oracle and OpenAI.
Silicon Valley needs access to debt to meet its goals for AI-related
spending, but so far, Wall Street is largely giving a blank check for
the AI ambitions of the most creditworthy tech companies.

 AGENTS CAN'T CHOOSE BETWEEN STRUCTURE AND FLEXIBILITY (8 MINUTE READ)
[18]

 Deciding between Python and Markdown for agent specification involves
balancing structure with flexibility. Code-maximalism enforces
reliability but lacks adaptability, while Markdown-maximalism allows
for creativity but risks errors. Successful agent architectures use a
hybrid approach, leveraging Markdown for intent and code for structure
to combine flexibility with reliable execution.

⚡

QUICK LINKS

 THE AGENT HARNESS IS A SHELL. (SPONSOR) [19]

 We've cheated by giving agents one powerful tool from 1979 and called
it a day. Bash, MCP, and skills we got the whole stack wrong.

Here's why. [19]

 AI SUMMARIES IN GMAIL (1 MINUTE READ) [20]

 Google added AI Overviews to Gmail for Workspace users, enabling
natural language queries that return summarized answers across emails
without opening threads.

 MICROSOFT TO INVEST $1.8B IN AUSTRALIA TO EXPAND AI, CLOUD, AND
DIGITAL INFRASTRUCTURE (2 MINUTE READ) [21]

 Microsoft is investing $1.8 billion to significantly expand its cloud
computing and artificial intelligence infrastructure across Australia.

 BECOME A CURATOR FOR TLDR AI (3-5 HRS/WEEK, FULLY REMOTE) [22]

 We're looking for an engineer/researcher at a major AI lab or startup
to help write for 1M+ subscribers. Curators have been invited to
Google I/O and OpenAI DevDay, scouted for Tier 1 VCs, and get early
access to unreleased TLDR products. Learn more [23].

 WHITE HOUSE ACCUSES CHINA OF INDUSTRIAL-SCALE AI MODEL DISTILLATION,
COMMITS TO INTELLIGENCE SHARING WITH OPENAI, ANTHROPIC, GOOGLE (11
MINUTE READ) [24]

 Model distillation is a technique that can be used to train cheaper
models that approximate the original's capabilities at a fraction of
the cost.

 OLMOEARTH EMBEDDINGS EXPORT (8 MINUTE READ) [25]

 AI2 introduced embedding exports in OlmoEarth Studio, allowing users
to generate and share compact Earth observation representations for
downstream tasks like search and segmentation.

Love TLDR? Tell your friends and get rewards!

 Share your referral link below with friends to get free TLDR swag!

 https://refer.tldr.tech/5a192f5c/2 [26]

		 Track your referrals here. [27]

Want to advertise in TLDR? 📰

 If your company is interested in reaching an audience of AI
professionals and decision makers, you may want to ADVERTISE WITH US
[28].

Want to work at TLDR? 💼

 APPLY HERE [29], CREATE YOUR OWN ROLE [30] or send a friend's resume
to jobs@tldr.tech and get $1k if we hire them! TLDR is one of INC.'S
BEST BOOTSTRAPPED BUSINESSES [31] of 2025.

 If you have any comments or feedback, just respond to this email!

Thanks for reading,
Andrew Tan [32], Ali Aminian [33], & Jacob Turner [34]

 Manage your subscriptions [35] to our other newsletters on tech,
startups, and programming. Or if TLDR AI isn't for you, please
unsubscribe [36].

Links:
------
[1] https://tldr.tech/ai
[2] https://advertise.tldr.tech/
[3] https://a.tldrnewsletter.com/web-version?ep=1&lc=97c90d72-3e48-11f1-ab0b-7fc89b18583c&p=7018eb02-3fcf-11f1-9317-11254792d9bb&pt=campaign&t=1777037643&s=dac7cd705085a7fe2c3d3597c490635ae154bd07113d722bfb6fbc98c793d8d4
[4] https://www.crusoe.ai/cloud/managed-inference
[5] https://links.tldrnewsletter.com/igKeLA
[6] https://links.tldrnewsletter.com/B2Awl5
[7] https://techfundingnews.com/deepseek-20b-valuation-tencent-alibaba-first-funding-round/
[8] https://finance.yahoo.com/markets/stocks/articles/anthropic-just-overtook-openai-1-155312239.html
[9] https://links.tldrnewsletter.com/ZehF5z
[10] https://12gramsofcarbon.com/p/agentics-configuring-agents-is-still
[11] https://turbopuffer.com/
[12] https://turbopuffer.com/customers/cursor
[13] https://www.anthropic.com/engineering/april-23-postmortem
[14] https://links.tldrnewsletter.com/MHVIXA
[15] https://github.com/amazon-science/expert-upcycling
[16] https://links.tldrnewsletter.com/OhxLfh
[17] https://links.tldrnewsletter.com/bSJjPc
[18] https://frontierai.substack.com/p/agents-cant-choose-between-structure
[19] https://inference.sh/blog/opinions/harness-is-a-shell
[20] https://techcrunch.com/2026/04/22/ai-overviews-are-coming-to-your-gmail-at-work/
[21] https://links.tldrnewsletter.com/r8W9wN
[22] https://jobs.ashbyhq.com/tldr.tech/038c4419-5b48-4279-a75e-6f7a0afdb240
[23] https://jobs.ashbyhq.com/tldr.tech/038c4419-5b48-4279-a75e-6f7a0afdb240
[24] https://thenextweb.com/news/us-white-house-ai-model-distillation-china-theft
[25] https://allenai.org/blog/olmoearth-embeddings
[26] https://refer.tldr.tech/5a192f5c/2
[27] https://hub.sparklp.co/sub_8f6b153ee058/2

# TLDR AI 2026-04-24

## HEADLINES & LAUNCHES

### GPT 5.5

OpenAI released GPT-5.5 with improved agentic reasoning, tool use, and efficiency, matching prior latency while increasing performance across coding and knowledge tasks.

### DeepSeek Unveils Flagship AI Model a Year After Breakthrough

DeepSeek has rolled out new preview versions of its new flagship AI model. The V4 Flash and V4 Pro series comes with architecture upgrades and optimization improvements. It has a 1-million-token context window. While the service capacity for the V4 Pro series is extremely limited due to a computing crunch, DeepSeek expects pricing for the model to drop once Huawei releases its Ascend 950-powered computing clusters.

### Tencent, Alibaba to Back DeepSeek at $20B+ Valuation

DeepSeek is in talks for its first funding round at a $20 billion valuation, with Tencent and Alibaba interested. Tencent is seeking a 20% stake, but DeepSeek doesn't want to lose that much control. The valuation surged from $10 billion to $20 billion in days, illustrating significant investor interest.

### Anthropic Just Overtook OpenAI with $1 Trillion Valuation

Anthropic reached a $1 trillion valuation on Forge Global, surpassing OpenAI's $880 billion. The spike in Anthropic's value stems from scarce available shares and growing demand due to partnerships and increased use of its Claude Code tool. Current investor interest highlights Anthropic's position as a leading AI opportunity.

## DEEP DIVES & ANALYSIS

### Training for Accuracy in Search LLMs

Perplexity detailed a staged post-training approach that teaches models tool use, evidence gathering, and structured evaluation to improve accuracy in search-based AI systems.

### Agentics: AI Enablement Requires Managed Agent Runtimes

Enterprises need managed agent runtimes for effective AI integration, avoiding the complexities of manual configuration. Current solutions like Claude Code are unintuitive, leading to inefficient setups and potential security risks. Companies like Ramp and Stripe have developed custom solutions, but managed services are scarce, creating opportunities for new entrants in the market.

## ENGINEERING & RESEARCH

### An Update on Recent Claude Code Quality Reports

Anthropic listened to feedback that Claude's responses had worsened, and it has resolved the issue as of April 20. The startup traced the problem to three separate changes that affected Claude Code, the Claude Agent SDK, and Claude Cowork. The API was not impacted. This post looks at how Anthropic investigated and fixed the problem and the commitments the company has made to ensure similar issues are much less likely to happen again.

### OpenAI Privacy Filter Model

OpenAI released a lightweight open-weight model for detecting and redacting PII in text, designed for fast, local, context-aware privacy filtering workflows.

### Expert Upcycling

Amazon researchers released a method to expand Mixture-of-Experts models mid-training by duplicating and specializing experts, improving performance without increasing inference cost.

## MISCELLANEOUS

### AI Coding Firm Cognition in Funding Talks at $25 Billion Value

Cognition AI is in early talks to raise a round of funding that would more than double its valuation to $25 billion. The talks are ongoing and the terms could change. Cognition uses AI to streamline the process of writing and debugging code, with a focus on selling to businesses. Its flagship product, Devin, is being used by companies like Anduril and Microsoft.

### Oracle's Deluge of AI Debt Pushes Wall Street to the Limit

Oracle's $300 billion megadeal with OpenAI is testing the limit of Wall Street's appetite for debt tied to the datacenter boom. Banks have struggled for months to spread the risk of the billions of dollars in loans they made to build data centers leased to Oracle in Texas and Wisconsin. Bank balance sheets are now clogged, constraining the financing prospects of future projects tied to Oracle and OpenAI. Silicon Valley needs access to debt to meet its goals for AI-related spending, but so far, Wall Street is largely giving a blank check for the AI ambitions of the most creditworthy tech companies.

### Agents Can't Choose Between Structure and Flexibility

Deciding between Python and Markdown for agent specification involves balancing structure with flexibility. Code-maximalism enforces reliability but lacks adaptability, while Markdown-maximalism allows for creativity but risks errors. Successful agent architectures use a hybrid approach, leveraging Markdown for intent and code for structure to combine flexibility with reliable execution.

### AI Summaries in Gmail

Google added AI Overviews to Gmail for Workspace users, enabling natural language queries that return summarized answers across emails without opening threads.

### Microsoft to Invest $1.8B in Australia to Expand AI, Cloud, and Digital Infrastructure

Microsoft is investing $1.8 billion to significantly expand its cloud computing and artificial intelligence infrastructure across Australia.

### White House Accuses China of Industrial-Scale AI Model Distillation, Commits to Intelligence Sharing with OpenAI, Anthropic, Google

Model distillation is a technique that can be used to train cheaper models that approximate the original's capabilities at a fraction of the cost.

### OlmoEarth Embeddings Export

AI2 introduced embedding exports in OlmoEarth Studio, allowing users to generate and share compact Earth observation representations for downstream tasks like search and segmentation.

Tu es l'extracteur de contenu de Breviat. On te fournit le contenu Markdown nettoyé d'une newsletter.

Ta mission : produire une version PROPRE du contenu en supprimant tout ce qui n'est pas de l'information utile au lecteur. Tu es un FILTRE, pas un résumeur.

À RETIRER :
- Publicités, encarts sponsors, mentions "sponsorisé par X", "ad", "présenté par"
- Intros vides : formules de bienvenue, météo de l'humeur de l'auteur, anecdotes personnelles non liées au contenu
- Appels à l'action marketing : s'abonner à la newsletter, parrainer un ami, "follow us on Twitter", "join our Discord"
- Signatures, mentions légales, adresses postales, "view in browser", "unsubscribe"
- Boutons / CTAs / "cliquez ici" / "lire la suite" sans contenu derrière
- Promotions d'autres produits / événements / formations payantes de l'auteur ou de tiers
- Encarts récurrents type "Read of the day" ou "Quote of the day" sans valeur informationnelle propre

À CONSERVER (intégralement, sans résumer ni reformuler) :
- Toutes les annonces, news, analyses, commentaires factuels
- Les chiffres, dates, noms d'entreprises, citations
- Les explications techniques
- Les liens vers des sources réelles (annonces officielles, papers, articles cités)
- La structure (titres, sous-titres, listes)

RÈGLES :
- Ne reformule pas. Garde la formulation d'origine.
- Ne résume pas, ne condense pas. Si une section fait 200 mots et est utile, garde 200 mots.
- N'ajoute aucun contenu (pas de titres ni de transitions de ton cru).
- Ne fabrique aucune URL. Garde celles d'origine, ou retire-les.
- Si la newsletter entière est de la pub / promo / contenu inutile, sors un Markdown vide (rien d'autre).

Sortie : UNIQUEMENT le Markdown nettoyé, sans préambule ni commentaire sur ton travail.