HashNode

🚨 DeepSeek V3.1 Just Dropped With 6x Cheaper Output Costs Than GPT-5 🚨

While everyone was watching OpenAI and Anthropic, Chinese startup DeepSeek AI just released V3.1 - a hybrid model that's forcing the industry to reconsider pricing.

Here's why this matters:

🧠 Revolutionary Dual-Mode Architecture 🧠
First AI model with "Think" and "Non-Think" modes in one system:
• Non-Think: Lightning-fast responses for quick queries
• Think: Deep step-by-step reasoning for complex problems
• Toggle with a single "DeepThink" button
• Cuts resource costs by 50% vs running separate models

⚡ Crushing Performance Benchmarks ⚡
The numbers don't lie:
• SWE-Bench Verified: 66.0% (vs 45.4% previous version)
• Terminal-Bench: 31.3% (vs 13.3% before)
• AIME 2025: 88.4% in Think mode
• 671B parameters, 128K context window

💰 Pricing That Breaks the Industry 💰
DeepSeek: $0.56 input / $1.68 output per million tokens
GPT-5: $1.25 input / $10.00 output per million tokens
That's 6x cheaper output costs while maintaining competitive performance.

🔓 Open-Source Under MIT License 🔓
Free weights on Hugging Face, full commercial use allowed. DeepSeek is democratizing advanced AI like DJI did for drones.

🤖 Built for AI Agents 🤖
Enhanced tool calling, JSON function support, and multi-step workflows make this perfect for autonomous coding, research, and automation tasks.

The hybrid approach eliminates the need for multiple AI subscriptions. One model, two modes, fraction of the cost.

Available now at on Deepseek Chat and API.

This pricing disruption is forcing major players to reconsider their strategies.

Will this force OpenAI, Anthropic and Google to slash their prices?

👇👇 Benchmarks & Links 👇👇

Can you Solve the Question?

AI Is Killing Coding, agree or not?