--- title: Jev splits the internet, DeepSeek shrinks its cache 4x description: TypeSafe's Jev model dominated today's discourse, praised by some as a genuinely new agent primitive and torn apart by others over its compaction claims and... canonical: https://daily.dev/posts/jev-splits-the-internet-deepseek-shrinks-its-cache-4x-ic2dfvozs twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Jev splits the internet, DeepSeek shrinks its cache 4x | daily.dev og:description: TypeSafe's Jev model dominated today's discourse, praised by some as a genuinely new agent primitive and torn apart by others over its compaction claims and... og:url: https://daily.dev/posts/jev-splits-the-internet-deepseek-shrinks-its-cache-4x-ic2dfvozs og:image: https://api.daily.dev/og/posts/IC2DfVoZS.png og:image:alt: Jev splits the internet, DeepSeek shrinks its cache 4x og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Jev splits the internet, DeepSeek shrinks its cache 4x **[Agentic Digest](https://daily.dev/sources/agents_digest)** · 4 min read · 2 upvotes · 0 comments ## Summary TypeSafe's Jev model dominated today's discourse, praised by some as a genuinely new agent primitive and torn apart by others over its compaction claims and shaky benchmarks. DeepSeek quietly shipped V4.1 Flash with a 4x KV cache reduction that beats its own bigger model on agentic benchmarks. A zero-click RCE in Claude Code, Codex, Gemini CLI, and Copilot's plugin verification got patched by some vendors and ignored by others. Anthropic put Accenture, its own commercial partner, in charge of red-teaming its models, and the market reaction says a lot about how that landed. ## Content **TLDR:** TypeSafe's Jev model dominated today's discourse, praised by some as a genuinely new agent primitive and torn apart by others over its compaction claims and shaky benchmarks. DeepSeek quietly shipped V4.1 Flash with a 4x KV cache reduction that beats its own bigger model on agentic benchmarks. A zero-click RCE in Claude Code, Codex, Gemini CLI, and Copilot's plugin verification got patched by some vendors and ignored by others. Anthropic put Accenture, its own commercial partner, in charge of red-teaming its models, and the market reaction says a lot about how that landed. --- ## Jev, a 'System One' judgment model, splits developers between hype and teardown TypeSafe's Jev returns typed judgments instead of text, and Harrison Chase says it sparked more internal demos at LangChain than any model launch he's seen. But Theo's breakdown of Jev's compaction pitch is brutal: deciding what context to drop line-by-line, with no visibility into tool call results or encrypted reasoning traces, risks making agents dumber and actively increases cache-rewrite costs. A separate teardown found Jev's headline '193.6x faster, 244.6x cheaper' claims were measured from vendor laptops with no reported accuracy numbers against ground truth. Structured output guarantees validity, not correctness - worth remembering before wiring this into anything that matters. [Read more](https://daily.dev/feed-by-ids?id=fEIRV2rLx&id=FqoujnFcB&id=a0bfdwslu&id=7n3mDfaV4&id=VHhtg3LDG&id=BPY0i1Zlx) ## DeepSeek V4.1 Flash cuts KV cache 4x with a split encoder-decoder design DeepSeek's new architecture, Causal Encoder-Decoder, splits a 40-layer model so 20 layers handle prefill and 20 handle decode, pulling KV cache straight from encoder outputs instead of recomputing it. Combined with cross-layer sharing and 4-bit storage, cache size drops to roughly 890 bytes per token, a quarter of V4-Flash's footprint. At max reasoning effort it reportedly beats the bigger V4-Pro on agentic benchmarks like Terminal-Bench 2.1, though it burns more tokens getting there, which matters if you're latency-sensitive rather than throughput-bound. [Read more](https://daily.dev/posts/2xbydASIa) ## Plugin4Shell RCE hits every major coding agent, and half the vendors haven't patched it Researchers found that Claude Code, Codex, Gemini CLI, and Copilot all trust a Git SHA without verifying Git actually checked out that commit, letting an attacker swap in malicious plugin code with zero user interaction. Since these agents run with full developer access, a successful exploit can hand over source code, credentials, and CI/CD pipelines. Anthropic and OpenAI patched it, Google killed Gemini CLI instead of fixing it, and GitHub hasn't shipped a fix at all - so if you're on Copilot with third-party plugin marketplaces, you're exposed right now. [Read more](https://daily.dev/posts/KmY9J53Rq) ## Anthropic hires its own commercial partner Accenture to red-team its models Anthropic named Accenture's Faculty division as its first embedded AI evaluator, despite running a joint 'Accenture Anthropic Business Group' together and both companies committing $1 billion over five years to the arrangement. Most people expected this role to go to an independent nonprofit like METR or Apollo Research. Accenture's stock jumped 8% after-hours, which tells you how the market read it - and Anthropic is reportedly still negotiating with METR separately for something that might actually be independent. [Read more](https://daily.dev/posts/RdOvYO4ZB) --- ## Also notable - **Shopify drops React Native for native Swift/Kotlin, citing AI cross-platform porting:** Shopify rebuilt its Shop app in 12 weeks with native code, betting that coding agents now port features between platforms automatically, making cross-platform frameworks unnecessary for teams with that tooling. [Read more](https://daily.dev/posts/ob7qFAvyG) - **Vercel's gateway data shows open-weight tokens crossing 56% share, but Anthropic still takes 64 cents of every dollar:** Open-weight models went from 7% to 56% of tokens on Vercel's AI Gateway since December, yet capture only 14% of spend, while Anthropic's share of revenue has never dropped below 61% since December. [Read more](https://daily.dev/posts/3l2Uwb92D) - **PayPal's WebMCP checkout cuts agent compute cost 4.6x and doubles speed:** PayPal's case study shows three structured WebMCP tool calls replacing 30+ manual UI interaction steps, delivering roughly 4.6x cheaper inference and 2x faster median checkout completion for AI agents. [Read more](https://daily.dev/posts/weEUdfSfe) - **Manus seeks $4B valuation five months after Meta's blocked $2B acquisition:** The Chinese AI agent startup is in talks to raise $500M at a $4B valuation and is weighing a Hong Kong IPO, after Beijing blocked its Meta acquisition over export control concerns. [Read more](https://daily.dev/posts/hbjkdraNY) - **Anthropic runs a physical wet lab while loosening some biosafety guardrails on Claude:** Anthropic confirmed it operates a Bay Area wet lab for biology experiments backing its Claude Science product used by Novo Nordisk and Genentech, even as it publicly warns about AI biological weapon risks. [Read more](https://daily.dev/posts/Sa44y8ltp) ## Similar posts on daily.dev - [Why not use DeepSeek Flash for everything? · Los Techies](https://daily.dev/posts/why-not-use-deepseek-flash-for-everything-los-techies-sq77ike6a) · LosTechies · 1 upvotes · 1 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#anthropic](https://daily.dev/tags/anthropic), [#ai-inference](https://daily.dev/tags/ai-inference), [#deepseek](https://daily.dev/tags/deepseek) [View this post on daily.dev](https://daily.dev/posts/jev-splits-the-internet-deepseek-shrinks-its-cache-4x-ic2dfvozs) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/jev-splits-the-internet-deepseek-shrinks-its-cache-4x-ic2dfvozs","headline":"Jev splits the internet, DeepSeek shrinks its cache 4x","text":"TypeSafe's Jev model dominated today's discourse, praised by some as a genuinely new agent primitive and torn apart by others over its compaction claims and shaky benchmarks. DeepSeek quietly shipped V4.1 Flash with a 4x KV cache reduction that beats its own bigger model on agentic benchmarks. A zero-click RCE in Claude Code, Codex, Gemini CLI, and Copilot's plugin verification got patched by some vendors and ignored by others. Anthropic put Accenture, its own commercial partner, in charge of red-teaming its models, and the market reaction says a lot about how that landed.","url":"https://daily.dev/posts/jev-splits-the-internet-deepseek-shrinks-its-cache-4x-ic2dfvozs","datePublished":"2026-09-19T04:17:44.691Z","dateModified":"2026-09-19T04:18:05.861Z","author":{"@type":"Organization","name":"Agentic Digest","logo":"https://media.daily.dev/image/upload/s--V91DY4ls--/f_auto,q_auto/v1772617267/logos/agents_digest","url":"https://daily.dev/sources/agents_digest"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"isPartOf":{"@type":"WebPage","url":"https://daily.dev/sources/agents_digest","name":"Agentic Digest"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Agentic Digest","item":"https://daily.dev/sources/agents_digest"},{"@type":"ListItem","position":3,"name":"Jev splits the internet, DeepSeek shrinks its cache 4x"}]} ```