You’ll get a concise rundown of the latest AI‑dev news: SpaceX’s planned acquisition of Cursor, DeepSeek’s V4 models, a Claude Code source leak, Windsurf’s Devin integration, and Cognition’s push for a $25 billion valuation.
No noise. Just what matters. Here are the five stories shaping how developers build with AI.
1. SpaceX Buys Cursor for $60 Billion
April 21, 2022
Elon Musk's SpaceX secured the right to acquire Cursor, the AI-powered IDE used by millions of developers, for $60 billion later this year. The alternative structure in the deal: pay $10 billion just for the collaboration. Cursor's projected annual recurring revenue stands at $6 billion by end of 2026, a number that makes the acquisition price easier to understand in context.
Cursor was on track to close a $2 billion private funding round when SpaceX stepped in with the offer, and the startup chose to halt those discussions. The deal is being delayed until after SpaceX's IPO this summer, largely because the company wants to avoid updating its financial filings before the listing and prefers to finance the purchase using newly public stock.
The deeper story is what this means for the model layer. Cursor currently runs on Claude and GPT models, meaning Anthropic and OpenAI are powering the product that SpaceX is about to own. If the acquisition closes, that arrangement is unlikely to survive. xAI's Grok becomes the obvious replacement, which would make Cursor the first major developer IDE to cut ties with both dominant AI labs simultaneously.
The ultimate shortcut to flawless AI results
Stop wasting time guessing prompts. Get consistent, professional AI results right from the first try, every time.
Hint: For developers, the practical question is simple: how much does your workflow depend on Cursor's current model quality, and what happens if that changes overnight?
2. DeepSeek V4 Dropped: Frontier AI at 7x Lower Cost
April 24, 2026
DeepSeek released two new models, V4-Pro and V4-Flash, both open-source under the MIT license and both supporting a 1 million token context window by default. V4-Pro is the flagship with 1.6 trillion total parameters and 49 billion active per token. V4-Flash is the cost-optimized variant at 284 billion total parameters and 13 billion active.
The benchmark numbers are hard to ignore. V4-Pro scores 80.6% on SWE-bench Verified, the standard real-world software engineering benchmark. Claude Opus 4.7 scores 80.8%. The gap is smaller than the margin of error on most benchmark runs. On LiveCodeBench, V4-Pro scores 93.5% against Claude's 88.8%. On Terminal-Bench, 67.9% against Claude's 65.4%.
Model
SWE-bench
LiveCodeBench
Terminal-Bench
Output price / 1M tokens
DeepSeek V4-Pro
80.6%
93.5%
67.9%
$3.48
Claude Opus 4.7
80.8%
88.8%
65.4%
$25.00
DeepSeek V4-Flash
—
—
—
$0.28
For teams running high-volume coding workloads, the math is immediate. Switching saves tens of thousands of dollars per month at scale. The migration path requires almost no effort. DeepSeek's API is fully OpenAI-compatible and also supports the Anthropic API format. Both models are already integrated into Claude Code, OpenClaw and OpenCode as drop-in alternatives.
Hint: Switching is literally one line of code. Update the model string to `deepseek-v4-pro` or `deepseek-v4-flash` and keep everything else identical.
3. Claude Code Source Leaked: 512,000 Lines of Anthropic Secrets
March 31, 2026
A single forgotten .map file in an npm package exposed the entire Claude Code codebase. Security researcher Chaofan Shou spotted the file in version 2.1.88 and posted about it on X. The post reached 28.8 million views. The GitHub mirror accumulated 84,000 stars and 82,000 forks before Anthropic could issue DMCA takedowns. Anthropic confirmed the incident was caused by human error in the release packaging process, not a security breach.
The leaked codebase contained 1,906 TypeScript files across roughly 512,000 lines of code. The most notable findings from community analysis:
Agent swarms — Claude Code can spawn sub-agents with restricted toolsets running in isolated contexts, gated behind a feature flag
KAIROS — an unreleased always-on background agent mode with a nightly memory distillation feature
Anti-distillation — a mechanism that injects fake tool definitions into API requests to poison the training data of anyone recording Claude Code traffic to train a competing model
Undercover Mode — strips all references to Anthropic internals when Claude Code is used in external repositories, meaning AI-authored commits carry no indication that an AI was involved
A total of 44 compile-time feature flags were found in the codebase, covering capabilities that are fully built but not yet shipped externally. Because Bun's compile-time dead code elimination removes flagged-off modules entirely, none of these features exist in the binary that ships to users.
Heads up: A separate supply chain attack hit the same day. Anyone who ran npm install between 00:21 and 03:29 UTC on March 31 should downgrade to a safe version and rotate all secrets immediately.
4. Devin is Now Inside Windsurf: One-Click Autonomous Agent in Your IDE
April 15, 2026
Windsurf version 2.0.44 shipped Devin Cloud as a native feature inside the editor, included in every paid self-serve plan. The integration works as follows: you identify a task that can be cleanly scoped, click to delegate, and Devin picks it up in its own virtual machine. It writes the code, runs the tests, and opens a pull request. Progress is visible in a Kanban-style panel inside the editor without switching context or losing your flow.
The same update also shipped a set of significant additions alongside the Devin integration:
Claude Opus 4.7 — now available as a model inside Windsurf
Parallel multi-agent sessions — via Git worktrees, previously exclusive to Claude Code's agent teams feature
Plan Mode — Cascade now analyzes the full scope of a task before writing any code, reducing wasted iterations
New pricing — Windsurf Pro moved from $15 to $20 per month, matching Cursor's Pro tier
Devin Cloud is disabled by default for enterprise accounts. Enterprise admins need to enable it explicitly in organization settings if they have already purchased the Cognition Platform.
Hint: Devin delegation works best for clearly scoped tasks with verifiable outputs. Handing off anything that requires frequent judgment calls mid-task does not work reliably yet.
5. Cognition Eyes $25 Billion Valuation: From Meme to Market Leader
April 23, 2026
Cognition AI, the company behind Devin, is in early discussions to raise hundreds of millions of dollars at a valuation of approximately $25 billion. That would more than double its valuation from $10.2 billion, set just seven months earlier when Founders Fund led a $400 million round alongside Lux Capital, 8VC and others.
The growth trajectory behind the number is real. Devin's annual recurring revenue grew from $1 million in September 2024 to $73 million by June 2025, a 73-fold increase in nine months. After acquiring Windsurf in July 2025, Cognition's combined enterprise ARR more than doubled. The customer list now includes Goldman Sachs, Citi, Microsoft, Dell, Cisco, Palantir, Ramp and Nubank.
Metric
Then
Now
ARR
$1M (Sep 2024)
$73M (Jun 2025)
Valuation
$4B (Mar 2025)
$25B (target)
Task success rate
~15% (2024 demo)
Production at Goldman, Citi, Microsoft
In 2024, Devin launched to public mockery. Independent tests showed it completing roughly 15% of assigned tasks successfully. Two years later the same product is running production engineering work for some of the largest financial institutions in the world.
Hint: The valuation jump was triggered directly by the SpaceX-Cursor deal. When Cursor left the independent market, investor interest flooded into Cognition overnight.
