DeepSeek V4 Flash scored 82.7 on Terminal-Bench 2.1, a benchmark for real terminal and command-line tasks, ahead of Anthropic’s Claude Fable 5 at 80.5, while GPT-5.6 Sol topped the chart at 85.8 and Claude Sonnet 5 trailed at 74.5. V4 Flash is priced at $0.14 per million input tokens and $0.28 per million output tokens, compared to Fable 5’s $10 and $50 for the same, making it roughly 99% less expensive. On Artificial Analysis’s Intelligence Index, which tracks average dollar cost per task, DeepSeek V4 Flash costs $0.03 per task, versus Kimi K3 at $0.86, GPT-5.6 Sol at $1.86, Opus 5 at $2.34, and Fable 5 at $3.15, making DeepSeek roughly 105 times cheaper on a like-for-like basis. This follows DeepSeek’s pattern from R1’s January 2025 launch, which matched OpenAI’s o1 on reasoning benchmarks at a fraction of the cost. DeepSeek has since released models under permissive licenses, including DeepSeek Math V2 and V3.2-Speciale, allowing developers to fine-tune and self-host. V4 Flash is not the top scorer but offers an