🧠 Google dropped "Deep Thinking" in Gemini 3

Gemini just got a serious upgrade. Google’s new “Deep Think” mode is designed to handle complex reasoning tasks — the kind that require patience, not just pattern matching.

Claude is opening up. Anthropic is offering deep discounts (up to 75%) and nonprofit-focused integrations that make adoption easier for real-world orgs.

And DeepSeek? Quietly releasing a model that claims GPT‑5‑level performance — with a far smaller training budget.

We’re also switching up how we bring you the newsletter.
Starting today, it’s:

Three big updates. Two tools to try. One workflow you’ll actually use.

Let’s get into it.

3 Big updates

1) Gemini 3 “Deep Think” is live for AI Ultra subscribers

Google rolled out Deep Think, a higher-effort reasoning mode in Gemini 3 for AI Ultra users. When turned on, Gemini slows down to reason through steps, keeps intermediate states, and checks itself before answering.

You get it inside the Gemini mobile app if you’re an AI Ultra user ($249.99/month in the U.S.).

Why you should care: This is the first real mass-market rollout of deliberate reasoning mode. It nudges us toward a future where we ask AI to solve budgets, logical proofs, planning workflows, and step-wise models — not just text rewriting.

2) Anthropic’s Claude for Nonprofits (up to 75% off + integrations)

Anthropic launched a nonprofit program with discounts up to 75%, plus connectors for Blackbaud, Candid, and Benevity, and a free AI Fluency course to help teams adopt AI safely.

Why you should care: This widens access dramatically. Most nonprofits don’t have AI capacity — this reduces the friction of:

✔ cost
✔ onboarding
✔ workflow compatibility

If you advise nonprofits, this lets you show value without huge implementation burdens.

➡️ Current edge case:

The discount applies only to registered nonprofits — not social enterprises or mission-driven startups. Expect lobbying for expansion.

3) DeepSeek-V3.2 (and Speciale) focus on efficient reasoning

DeepSeek just released two new models — V3.2 and V3.2-Speciale — designed to solve complex math and logic problems, fast and cheap.

In testing, it beat GPT-4 on GSM8K reasoning and ARC-AGI-2 but fell short on writing tone and factual recall.

This matters because:

✔ If you build agent workflows (planners, solvers, code assistants), DeepSeek gives you cheap reasoning at scale

✔ If you write sales copy, research, or brand content, you’ll still get better output from Claude or GPT-4

2 Tools worth trying

Poetiq — An AI assistant built for deep, step-by-step reasoning — especially useful if you're working on logic-heavy tasks or benchmarking against advanced tests like
ARC-AGI-2.

Trynormal — An AI creative partner that studies your brand’s look and voice straight from your website, then helps you generate copy, visuals, and campaign assets that stay on-brand. Ideal for small teams that need to move quickly without compromising quality.

1 Workflow for the day

Let Gemini critique your video drafts
Gemini can analyze video + audio together. Upload your draft (demo, reel, lesson), then ask for a summary, pacing notes, clarity fixes, and a punch-list with timecodes.

Starter prompt:

Calling the community

Have a repeatable workflow you rely on? Reply with a short description + tools used. We’ll feature the best ones next week and credit you.

Until next time,
Kushank @digitalSamaritan