xAI vs Apple+OpenAI, Gemini 2.5’s big jump, Anthropic’s $10B play
From Siri’s rumored Gemini reboot to Perplexity’s 80% publisher split and ByteDance’s 512k-context model—what actually matters for builders this week.
TLDR; (30-sec)
- xAI sues Apple & OpenAI.
- Perplexity launches Comet Plus: 80% revenue share for publishers + $42.5M fund.
- Gemini 2.5 Flash Image ships
- Apple and Google in talks: Gemini could power a revamped Siri (report).
- Anthropic seeks up to $10B raise.
- ByteDance releases Seed-OSS-36B with 512k-token context (open weights).
- Grok chats indexed by Google; shared conversation links appear in search, raising privacy/safety concerns.
📰 This Week’s Big Stories
xAI sues Apple and OpenAI, alleging anticompetitive collusion
The suit claims Apple’s OpenAI tie-up and App Store practices unfairly throttle rivals like Grok—setting up a major AI antitrust fight.
Perplexity launches Comet Plus with 80% revenue share for publishers
New subscription plan gives media partners the lion’s share of fees and includes a $42.5M fund to kickstart participation, per company announcement.
Google debuts Gemini 2.5 Flash Image (“nano-banana”)
The new image model adds multi-image fusion, character consistency, precise natural-language edits, API access, and SynthID watermarking. (Google Developers Blog)
Apple reportedly in talks to use Google’s Gemini for Siri revamp
Bloomberg (via Reuters) reports early discussions to power a redesigned Siri with Gemini, pushing Alphabet and Apple shares higher on the day.
Anthropic is in talks to raise up to $10B
The Claude maker is reportedly pursuing a multibillion round amid heavy investor demand, signaling aggressive expansion plans.
💻 Quick Build Tips
LLM Context Management: cut cost, boost quality
A practical guide to shrinking prompts, chunking smartly, and caching context for faster, cheaper, more reliable agent runs. Bookmark-worthy. (eval.16x.engineer)
Product Fidelity > shipping faster
After v1, progress is all UX polish and detail work—the post argues speed matters less than dialing “feel” and fit. Great lens for PMs/designers.
7 mental models Jeff Bezos actually uses
From Regret Minimization to Type-1/Type-2 decisions—tight summaries of decision frameworks you can apply to AI product bets.
🛠️ Tools & Frameworks
Stitch (Google): prompt-to-UI (then export to Figma/code)
Generate mobile/web UIs from text or reference images, iterate in chat, and export—an AI shortcut from idea to front-end.
Chance AI – Curiosity Lens (visual agent)
“Point & understand” camera app focuses on meaning over shopping links—visual reasoning and context on what you see.
Cipher (OSS): memory layer for coding agents
Adds persistent memory to IDE/CLI agents (Cursor, Claude Code, Gemini CLI, etc.) via MCP—so sessions don’t forget prior fixes.
Async Server (OSS): ship issues → PRs via agents
“Claude Code + Linear + GitHub PR” style workflow: route tasks to an agent that drafts changes and opens actionable PRs.
AGENTS.md: a simple spec for guiding agents
Romain Huet announces AGENTS.md, a lightweight “README for agents”—use a single, open format to describe capabilities/constraints, improving reproducibility across tools.
⚡ Quick Bits
Meta signs a six-year, $10B+ Google Cloud deal
A major win for Google Cloud as Meta scales AI infrastructure; neither company has publicly confirmed details yet.
Thousands of Grok chats are now indexed on Google
Shared Grok conversation URLs are being crawled, exposing user chats in search results—some containing dangerous prompts and sensitive content, raising fresh privacy and safety concerns.
ByteDance open-sources Seed-OSS-36B with 512k context
The 36B-parameter model ships a native 512,000-token window for long-context reasoning and document processing—serious firepower for OSS stacks.
📌 Deep Dives
Decoder: Amazon’s agent playbook with David Luan
Why Amazon thinks solving agents (not just bigger models) is the next platform shift—and how “model factories” outcompete labs chasing benchmarks.
No, AI progress isn’t slowing (it just looks that way)
A sober case that media coverage ≠ progress: architectural shifts, not press cycles, will deliver the next step-function.
Are AI unicorns finally maturing?
CB Insights tracks higher “commercial maturity” scores among top AI startups in 2025—signs of real traction beyond hype.
Do you have too many AI products?
Practical framework to stop tool sprawl: standardize interfaces, consolidate overlapping use-cases, and measure ROI beyond pilots.
AI FOMO is inflating a risky bubble, investors warn
SPVs are surging as retail investors chase shares in hot AI names like OpenAI and Anthropic—often with high fees and opaque structures—prompting caution about a growing bubble.
That's a wrap for this week!
👉 Found this useful? Forward it to a developer friend
👉 Have a tool suggestion? Reply and let me know
👉 Want more? Follow me on [Twitter] | [LinkedIn]
Want to sponsor a future issue? [Contact us here]
Attia