DeepSeek Drops GPT Level Reasoning

Hello there! Ready to dive into another upgrading, Mind-boggling, and Value-filled Shift?

Today we have:

🤖 DeepSeek V3.2 drops frontier reasoning at bargain pricing

🎬 How to Generate Perfect Video Prompts in Seconds with Hedra Prompt Autocomplete

🎞️ Runway & Kling kick off an AI-video arms race

🔨Tools and Shifts you cannot miss

🤖 DeepSeek V3.2 drops frontier reasoning at bargain pricing

DeepSeek has released V3.2 and the high-compute V3.2-Speciale, two open-source reasoning models rivaling GPT-5 and Gemini 3 Pro. The release combines three technical breakthroughs spanning sparse attention, scaled RL, and agentic data synthesis. The result is frontier-level performance with drastically lower costs.

The Shift:

1. Sparse Attention - DeepSeek introduced DSA, a new sparse-attention system that cuts compute in long-context tasks while preserving accuracy. This improves efficiency across math, code, and reasoning benchmarks, also enabling stable scaling without prohibitive hardware demands.

2. RL Scaling - A reinforced learning framework with expanded post-training compute pushes V3.2 to GPT-5-level performance. The Speciale variant even surpasses GPT-5 and matches Gemini-3-Pro in advanced reasoning. It validated these gains with gold-medal scores in the 2025 IMO and IOI.

3. Price Advantage - DeepSeek priced V3.2 at $0.28 input and $0.42 output per million tokens, undercutting Gemini 3 Pro, GPT-5.1, and Sonnet 4.5. This makes frontier reasoning accessible to startups and indie devs instead of only enterprise budgets. 

4. Open Weights - Both 685B-parameter models ship MIT-licensed with weights on Hugging Face. Developers can run, fine-tune, or self-host the entire stack without restrictions. 

DeepSeek is proving it’s not a one-off phenomenon; it’s a real competitor to U.S. frontier labs. For DTC, AI, and e-commerce operators, cheaper frontier reasoning means more automation, more agents, and more experimentation without API-bill anxiety. The pricing and openness will reshape competitive pressure across the entire model market.

^{TOGETHER WITH ADQUICK}

Modernize your marketing with AdQuick

AdQuick unlocks the benefits of Out Of Home (OOH) advertising in a way no one else has. Approaching the problem with eyes to performance, created for marketers with the engineering excellence you’ve come to expect for the internet.

Marketers agree OOH is one of the best ways for building brand awareness, reaching new customers, and reinforcing your brand message. It’s just been difficult to scale. But with AdQuick, you can easily plan, deploy and measure campaigns just as easily as digital ads, making them a no-brainer to add to your team’s toolbox.

Learn more now.

🎬 How to Generate Perfect Video Prompts in Seconds with Hedra Prompt Autocomplete

Hedra’s new Prompt Autocomplete turns rough ideas into production-ready prompts instantly. Type a few words, hit Tab, and it fills in camera moves, lighting, emotion, and dialogue for you.

1. Start with a messy idea: Type something like “skincare founder story ad” or “product unboxing in bathroom.”

2. Hit TAB to autocomplete: Hedra expands it into a fully optimized prompt with cinematic details, pacing, and tone baked in.

3. Generate multiple variants fast: Make 5 versions by changing one input word, like “luxury,” “UGC,” “comedic,” or “emotional,” then pick the best.

4. Use it for rapid creative testing: Turn each finished prompt into a new angle for ads, Reels, landing page hero videos, or brand storytelling.

✨ Bonus: Reply “Hedra Prompt Autocomplete” to get a code if you’re one of the first 500.

Try it here 

🎞️ Runway & Kling kick off an AI-video arms race

Runway’s Gen-4.5 and Kuaishou’s Kling O1 dropped almost back-to-back, pushing AI video into a new tier of realism and editability. Both models leapfrog previous benchmarks. The result is a clear acceleration toward studio-grade, brand-ready video generation.

The Shift:

Runway’s Gen-4.5

1. Benchmark Lead - Runway Gen-4.5 now tops the Artificial Analysis leaderboard, building on months of hype under codenames like Whisper Thunder. The model shows major gains in realism, fluidity, and motion consistency.

2. Cinematic Realism - Gen-4.5 improves physics, fluid dynamics, and human movement, keeping hair, fabric, and micro-details stable across frames. Runway claims outputs can appear indistinguishable from real footage, pushing AI video closer to Hollywood-level.

Kuaishou’s Kling O1

1. All-in-One Editing - Kling O1 accepts up to seven inputs and supports text-based edits like removing bystanders or shifting lighting. It preserves character identity while altering environments or styles, merging video creation, restyling, and granular editing.

2. Feature Dominance - Internal tests show Kling O1 outperforming Google Veo 3.1 and Runway Aleph in reference-based and editing tasks. With tools for camera motion, multi-subject control, and restyling, it introduces VFX-level precision for short-form content without requiring professional editing skills.

For brands, this means studio-quality product clips, instant reshoots, and frictionless creative iteration. The competition between Runway and Kling will rapidly expand what brands can produce without agencies or production budgets.

^{TOGETHER WITH THE CODE}

What 100K+ Engineers Read to Stay Ahead

Your GitHub stars won't save you if you're behind on tech trends.

That's why over 100K engineers read The Code to spot what's coming next.

• Get curated tech news, tools, and insights twice a week

• Learn about emerging trends you can leverage at work in just 10 mins

• Become the engineer who always knows what's next

Join 100k+ engineers

🔨AI Tools for the Shift

🏠 RoomX AI – Transform empty real-estate listings with realistic AI staging in just 30 seconds.

🎥 PXZ AI | Video Generator – Turn photos into dynamic videos using AI-powered camera movement.

✍️ Revenuesurf – Generate revenue-focused blog posts designed to convert and rank.

🍽️ PlatePhoto – Create professional, appetizing food photos in seconds to boost sales.

🎵 AnyMusic – Make AI-generated music lyrics and full songs for free with no signup required.

🚀Quick Shifts

🔄 OpenAI has taken a non-cash stake in Thrive Holdings, linked to Thrive Capital, an OpenAI investor, pairing AI tools and staff with IT and accounting firms, and potentially sharing returns.

🍎 Apple is reshuffling AI leadership after Siri delays. John Giannandrea steps back, staying as advisor until spring 2026. Amar Subramanya becomes VP of AI, reporting to Craig Federighi internally.

🎬 Runway launched  Gen-4.5, saying it improves prompt-following and “physical accuracy” for more cinematic, realistic text-to-video. It’s rolling out gradually, but still slips on object permanence and cause-and-effect timing.

🛒Amazon’s AI chatbot, Rufus, spiked on Black Friday. Purchases from Rufus sessions rose 100% vs 20% without. Adobe saw AI traffic +805% YoY, and AI-referred shoppers 38% likelier to buy.

🚗 Nvidia unveiled Alpamayo-R1, an open vision-language-action reasoning model for autonomous driving research, plus a Cosmos Cookbook on GitHub. It targets Level 4 autonomy by improving “common-sense” decisions using Cosmos-Reason foundations.

🤳AI Nugget of the Day

That’s all for today’s edition see you tomorrow as we track down and get you all that matters in the daily AI Shift!

If you loved this edition let us know how much:

Forward it to your pal to give them a daily dose of the shift so they can 👇