Top Stories
DeepSeek V4 Arrives — Open-Source, 1M Context, and Prices That Undercut Everyone
DeepSeek released its long-anticipated V4 model on April 24, dropping two variants: V4-Pro (1.6 trillion total parameters, 49 billion active) and V4-Flash (284 billion total, 13 billion active). Both support a 1-million-token context window and dual thinking/non-thinking modes.
The performance numbers put V4-Pro within striking distance of GPT-5.4 and Gemini 3.1 Pro on coding, math, and world knowledge benchmarks — and ahead of every other open-source model. But the pricing is the real story. V4-Pro costs $3.48 per million output tokens, compared to OpenAI’s $30 and Anthropic’s $25. V4-Flash goes even lower at $0.28 per million tokens. That’s roughly a 10x discount over frontier closed-source APIs for comparable quality.
The release also marks a milestone for U.S.-China chip decoupling: Huawei announced same-day support for V4 on its Ascend processors, with DeepSeek confirming the two companies worked closely on optimization. DeepSeek expects to push V4-Pro prices even lower as Huawei scales production of its Ascend 950 chips.
Sources: CNBC · Fortune · MIT Technology Review
OpenAI Ships GPT-5.5 Six Weeks After 5.4, Calls It ‘a New Class of Intelligence’
OpenAI released GPT-5.5 on April 23, rolling it out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex. The company calls it “the smartest and most intuitive to use model” yet — a claim it has now made three times in as many months.
OpenAI president Greg Brockman described GPT-5.5 as “a faster, sharper thinker for fewer tokens compared to 5.4,” with reduced hallucinations and stronger agentic capabilities. Instead of requiring careful step-by-step prompting, GPT-5.5 can take a “messy, multi-part task” and figure out the sequence of tools, checks, and decisions needed to complete it. Bank of New York CIO Leigh-Ann Russell highlighted “really impressive hallucination resistance” as critical for regulated industries.
The six-week turnaround from GPT-5.4 underscores how frontier labs have shifted from landmark releases to continuous, incremental updates. As Brockman himself acknowledged: “There are enough model releases that it’s probably getting hard to distinguish one from another.”
Sources: CNBC · Fortune · TechCrunch
Cohere Acquires Aleph Alpha in $20B Sovereign AI Merger
Canadian AI company Cohere announced on April 24 that it will acquire Germany’s Aleph Alpha, creating what both companies call a “transatlantic AI powerhouse” at a combined valuation of approximately $20 billion. Schwarz Group — the parent company of Lidl and Kaufland and a major Aleph Alpha shareholder — is investing $600 million into Cohere’s concurrent Series E round.
The merged entity will operate under the Cohere brand with dual headquarters in Canada and Germany, with Cohere shareholders receiving roughly 90% of the combined company. The strategic angle is sovereign AI: both companies have positioned themselves as trusted providers for governments and regulated industries that want AI infrastructure they control, rather than depending on U.S. hyperscalers.
The deal signals that AI sovereignty is becoming a serious commercial category, not just a policy talking point. European governments have been vocal about wanting alternatives to Google and OpenAI, and the Cohere-Aleph Alpha combination is a direct bet that enough institutional buyers share that concern to support a $20 billion business.
Sources: PitchBook · BusinessWire
Quick Hits
-
SpaceX secures option to buy Cursor for $60B: SpaceX struck a deal giving it the right to acquire AI coding startup Cursor for $60 billion later this year, or pay $10 billion for the partnership alone. The acquisition would happen after SpaceX’s planned summer IPO, combining Cursor’s code-generation tools with SpaceX’s Colossus training supercomputer. Microsoft had also looked at buying Cursor before the SpaceX deal materialized. TechCrunch · CNBC
-
Cognition AI seeks $25B valuation: The maker of AI coding agent Devin is in early talks to raise a new round that would more than double its valuation. The AI coding market is now drawing valuations that rival established SaaS companies — and neither Cursor nor Cognition are profitable yet. Bloomberg
-
Hinton tells UN conference to ‘apply the brakes’: Nobel laureate Geoffrey Hinton compared unregulated AI to “a very fast car with no steering wheel” at the Digital World Conference in Geneva. He estimated that “maybe one percent” of work on AI goes toward making it safer, and pressed for mandatory safety spending. UN News
-
Bipartisan AI transparency bill advances: Reps. Beyer (D-VA), Lawler (R-NY), and Jacobs (D-CA) introduced the AI Foundation Model Transparency Act (H.R. 8094), requiring developers of large AI models to disclose training data, capabilities, risks, and monitoring practices to the FTC and public. Fully open-source models are exempt. Congress.gov
Worth Watching
DeepSeek’s pricing is a strategic weapon, not just a feature. At $3.48 per million output tokens versus $25-30 for frontier U.S. models, DeepSeek V4-Pro creates an uncomfortable question for enterprise buyers: how much are you paying for the name on the invoice? The performance gap to GPT-5.4 is described as “three to six months” — but the price gap is an order of magnitude. If V4-Pro holds up under real workloads, the pricing pressure on OpenAI and Anthropic will be impossible to ignore, especially as Huawei scales the Ascend chips that make those prices viable.
The AI coding market just got very expensive. SpaceX valuing Cursor at $60 billion while Cognition seeks $25 billion suggests the market believes AI-assisted software development is a winner-take-all race. Both bets are enormous relative to revenue. If the coding copilot market consolidates as fast as the underlying models are improving — with Qwen 3.6-27B and DeepSeek V4 now matching frontier code performance — the question becomes whether these tools can maintain pricing power as the models underneath them commoditize.