SunBrief#69: Google and Anthropic Just Dropped Their Biggest AI Upgrades

Welcome to the SunBrief

Today in SunBrief 🌞

• Investing Shouldn’t Feel This Hard

• Anthropic Launches Claude Opus 4.6 while OpenAI Strikes Back with GPT-5.3-Codex

• Zuckerberg Testifies as Meta Faces Teen Safety Questions

• Stock Updates

• Three Ex-Google Engineers Charged With Stealing Chip Security Trade Secrets

• AI Highlights of the Week

• Too Important to Miss

Investing Shouldn’t Feel This Hard

The problem is too much noise.

Too many headlines. Too many opinions. Too many “experts” shouting different directions.

Market Munchies cuts through it — with simple, bite-sized investing insights and high-value opportunities delivered straight to your inbox. Curated, edited and published by Mode Mobile.

And right now, Market Munchies readers can claim FREE shares in Mode Mobile, one of the fastest-growing software companies in the U.S.

You don’t need to be a pro. You just need better access.

Anthropic Launches Claude Opus 4.6 while OpenAI Strikes Back with GPT-5.3-Codex

On Feb 19, 2026, Google launched Gemini 3.1 Pro, just after Anthropic’s Feb 17, 2026, release of Claude Sonnet 4.6, sharpening the matchup between Gemini’s technical reasoning push and Sonnet’s practical, human-aligned workflow focus.

Gemini 3.1 Pro Key Upgrades

• Advanced core reasoning – Smarter baseline for complex problem-solving when simple answers aren’t enough

• ARC-AGI-2 breakthrough – Verified 77.1% score, more than double Gemini 3 Pro’s reasoning performance

• Practical intelligence – Clear visual explanations and better synthesis of data into a single view

• Code-based creativity – Generates website-ready animated SVGs from text prompts (pure code, crisp at any scale, lighter than video)

• Broad rollout – Available across Gemini app, NotebookLM, AI Studio + Gemini API, Vertex AI, Gemini Enterprise, Gemini CLI, Android Studio, and Antigravity

Claude Sonnet 4.6 Key Upgrades

• Opus-level capability at Sonnet pricing – Near-frontier reasoning and coding while keeping Sonnet 4.5 pricing ($3 / $15 per million tokens)

• 1M-token context window (beta) – Handles entire codebases, long contracts, or multiple research papers in a single request with strong long-context reasoning

• Developer-preferred coding upgrade – In Claude Code testing, Sonnet 4.6 was preferred ~70% over Sonnet 4.5 and ~59% over Opus 4.5

• Computer use leap (OSWorld-verified) – Major gains in real software interaction across browsers, spreadsheets, and IDEs without special connectors or APIs

• Safer for real-world tool use – Stronger resistance to prompt injection than Sonnet 4.5, with safety performance comparable to Opus 4.6

Comparison Between Two: 

• Best use-case – Gemini 3.1 Pro excels at technical clarity and structured explanations; Sonnet 4.6 shines in real-world judgment and practical planning

• Complex strategy – Sonnet 4.6 delivers more politically realistic city recovery plans with clear tradeoffs and constraints; Gemini 3.1 Pro focuses more on urban design and zoning mechanics

• Fast money plan – Sonnet 4.6 produces execution-ready 60-day income plans with realistic outreach and risks; Gemini 3.1 Pro offers higher-leverage product strategy but slower traction

• Tone and messaging – Gemini 3.1 Pro sounds more natural and ready-to-send for polite declines; Sonnet 4.6 comes across warmer and more heartfelt

• Explaining LLM reasoning – Gemini 3.1 Pro provides more technical, conceptually satisfying explanations; Sonnet 4.6 frames “thinking” in a more human, intuitive way

Zuckerberg Testifies as Meta Faces Teen Safety Questions

Landmark case probes whether Instagram and other social apps are addictive and damaging to minors

Meta CEO Mark Zuckerberg testified in an LA Superior Court trial over claims that Meta’s platforms are addictive and harmful to teens, with internal documents raising concerns about parental controls, time-spent pressure, and youth safety features.

Key Points:

• Addiction and Vulnerability Claims: Trial testimony cited Meta research indicating parental supervision may not stop compulsive use, and teens facing trauma may be more prone to overuse.

• Time-Spent Growth Push: Plaintiffs presented a 2015 email chain showing Zuckerberg pushing to raise time spent by 12%, challenging earlier assertions that teams weren’t set usage-growth goals.

• Beauty Filters Under Fire: Zuckerberg was questioned on Instagram beauty filters after internal experts reportedly recommended banning them for teens.

• Under-13 Usage Evidence: Internal documents raised questions about the scale of underage users, including an estimate that millions of children under 13 had Instagram accounts.

• Age Verification Debate: Zuckerberg argued age checks are difficult and suggested device makers like Apple could play a larger role; Apple has rolled out age assurance tools for developers.

• High-Stakes Trial Context: TikTok and Snap settled before trial; YouTube and Meta are defending their platforms, while Meta’s lawyers argue the plaintiff’s mental health issues stem from her upbringing, not social media.

Why It Matters:
A jury verdict could reshape how social platforms design for minors, accelerating regulation, product restrictions, and potential settlements, setting a precedent for how courts weigh internal research against claims of teen harm.

Stock Updates

Three Ex-Google Engineers Charged With Stealing Chip Security Trade Secrets

DOJ alleges sensitive processor and cryptography files were exfiltrated and accessed from Iran

Federal prosecutors charged three former Google engineers with conspiracy and trade secret theft, saying confidential mobile security and cryptography files were sent to unauthorized locations, including links to Iran, raising national security concerns.

Key Points:

• Criminal Charges Filed: The defendants face trade secret theft counts (up to 10 years each) and an obstruction charge carrying up to 20 years.

• Alleged Exfiltration Method: Prosecutors say hundreds of files, including Google trade secrets, were moved to a third-party communications platform and copied onto personal and other employers’ devices.

• Iran Access Allegation: The DOJ claims trade secret materials were accessed while one defendant was in Iran, including photos of a work screen taken the night before traveling.

• Concealment Claims: Prosecutors allege the trio deleted files, destroyed records, and signed false affidavits denying they shared confidential information.

• Internal Detection: Google’s security systems reportedly flagged suspicious activity and revoked one defendant’s access in August 2023.

Why It Matters:
The case highlights how insider access to advanced semiconductor and cryptography IP can become a national-security risk and why companies are under pressure to tighten monitoring, compartmentalize sensitive systems, and treat “legitimate access” as a core attack surface.

AI Highlights of the Week

• Nvidia in Talks to Invest Up to $30B in OpenAI

  Nvidia is in talks to invest up to $30B in OpenAI, in a funding round that could value the startup at $730B pre-money, sources say.

  The potential investment is separate from the $100B infrastructure deal, not tied to deployment milestones, and still not final, with OpenAI also raising funds from Microsoft, Amazon, and other strategic investors.

• China’s Humanoid Robots Leap From Stumbles to Kung Fu Flips

  China’s humanoid robots stunned audiences at the Spring Festival Gala, showcasing kung fu moves, flips, and fluid choreography, a sharp jump from their viral stumbles just one year ago.

  Analysts say China now leads with 85% of global humanoid deployments, driven by integrated supply chains and lower costs, but long-term impact will depend on real-world capability, not flashy stunts.

• Gemini Now Lets You Create Music With AI

  The Gemini app now includes Lyria 3, an advanced AI music generation model that lets anyone create 30-second tracks using text or images, now rolling out in beta.

  With auto-generated lyrics, style and tempo control, and realistic, high-quality sound, Lyria 3 makes music creation fast, fun, and shareable right inside Gemini..

Too Important to Miss

Meta to Shut Down Messenger’s Standalone Website in April 2026

Meta Quest 3’s Flagship App Shifts From VR to Mobile

OpenAI and Paradigm launch EVMbench to test AI agents on smart contracts

India’s Sarvam Launches Indus AI Chat App to Take On Global Rivals

Last Week’s Poll Result

• Do you think AI reasoning models like Deep Think can meaningfully assist scientific research?

  Yes, they’ll accelerate discovery → 53.57%
  Maybe, as decision-support tools → 32.14%
  No, research still needs human intuition → 14.29%

• Will AI-driven cyberattacks accelerate a global cyber arms race?

  Maybe, but escalation will be gradual → 52.63%
  Yes, it’s already happening → 26.32%
  No, it’s being overstated → 21.05%

• Half of xAI’s co-founders are gone. Is that a red flag?

  Yes, that’s instability → 47.06%
  Maybe, depends on context → 41.18%
  No, restructuring is normal → 11.76%

Feedback

Know someone who may be interested?