DeepSeek V4: Open-Source Frontier AI at a Fraction of the Price

On April 24, 2026 — the same day OpenAI released GPT-5.5 — DeepSeek dropped V4. Two variants: V4-Pro and V4-Flash. Both with 1 million token context windows, 384K max output, and API pricing that undercuts every closed model at the frontier. V4-Flash starts at $0.14 per million input tokens. GPT-5.5 is $5.00. That’s a 35x cost difference for comparable capability.

V4 at a Glance

• Context window: 1,000,000 tokens
• Max output: 384,000 tokens (8x the V3 limit)
• V4-Flash pricing: $0.14/M input, $0.28/M output
• V4-Pro pricing: $1.74/M input, $3.48/M output
• Cache hits: 1/10th standard input price
• Architecture: Compressed Sparse Attention enabling efficient 1M context
• License: MIT — full commercial use, self-hosting, and modification permitted

Why Open Source at the Frontier Changes Everything

Every previous frontier AI release has come from a closed lab with a proprietary model. You use it via API, accept the pricing, and have no option to self-host or inspect the weights. DeepSeek V4’s MIT license breaks that model entirely. An enterprise with data sovereignty requirements can run V4-Pro on their own infrastructure, with no data leaving their network and no per-token API cost. A startup can fine-tune V4-Flash on their own data and ship it as part of their product. The capability ceiling for open-source AI just moved to within benchmark distance of the best closed models.