DeepSeek V4.1 Flash hands-on tests show strong capability at a fraction of GPT's cost

DeepSeek's newly open-sourced model, DeepSeek V4.1 Flash 1017, was put through a wave of independent hands-on tests around October 18, delivering impressive results across coding, 3D world generation, and video generation, with multiple testers reporting costs far below GPT 6.1 Sol at comparable or even better quality and speed.

Confirmed

• YouTube channel WorldofAI ran comprehensive benchmarks on the model using its in-house tool woaibench.ai, covering coding, 3D world generation, and more, highlighting its ultra-low API pricing.
• In a video generation comparison (relayed by @gaganghotra from a blogger's test), generating a rocket launch video: DeepSeek cost $0.78 and took 3 minutes, while GPT 6.1 Sol cost $6.10 and took 5 minutes — nearly 8x cheaper and faster.
• FabulousPollution10, an engineer at Nebius working on coding agents and evals, tested it on a self-built benchmark of 10 real-world work tasks (3 runs per task, verified by hidden tests): all 30 runs passed at the high tier, with per-task cost under 5 cents.
• Blogger 花叔 (relayed by AlchainHust) stress-tested the model in Claude Code via DeepSeek's official Anthropic-compatible endpoint: 500 million tokens for just 28 yuan, and shared 9 example prompts.
• 量子位 rounded up users' 3D generation tests: including a physics-accurate Tesla flying car (capable of takeoff and flips), a sunflower with bees, a rocket launch site, an electric motor animation, and a floating island fantasy scene; compared with GPT-6.1 Sol, costs differed by up to 26x, with the model dubbed an "accessible 3D studio for everyone."

Why it matters

• Results from multiple independent sources and different benchmarks corroborate each other: the model is highly usable on real tasks like coding agents and 3D/video generation, offered at orders-of-magnitude lower cost, putting direct competitive pressure on expensive closed-source APIs.

2026-10-18 ~ 2026-10-18 · 5 related posts

• DeepSeek video model test: 8x cheaper and faster than GPT 6.1 Sol — gaganghotra_ · 2026-10-18
• [source] DeepSeek V4.1 Flash 1017 aces 10-task coding benchmark at under 5 cents per task — Fabulous_Pollution10 · 2026-10-18
• [source] 500M tokens for $4: hands-on with DeepSeek V4.1 Flash inside Claude Code, 9 prompts released — AlchainHust · 2026-10-18
• DeepSeek V4.1 Flash becomes everyone's 3D studio at up to 26x lower cost — 量子位 · 2026-10-18
• [source] DeepSeek V4.1 Flash Fully Tested: Coding, 3D Generation and Ultra-Low API Pricing Benchmarked — WorldofAI · 2026-10-18