DeepSeek V4-Flash holds 1M context under MIT — 284B/13B-active MoE proves the Flash-tier-open-frontier convergence

DeepSeek's V4-Flash variant (284B total / 13B active parameters, 1M context, MIT license) holds production-tier capability at hyperscaler-routable scale. Combined with V4-Pro (1.6T total / 49B active, 80.6 SWE-Bench Verified, 90.1 GPQA Diamond), DeepSeek now ships the most operationally credible open-weight Pro/Flash split. The 1M context retention in Flash is the structural detail that erases the case for routing to Pro on long-document workloads.

The licensing detail is what closes the procurement bar. MIT — not Apache 2.0 with reciprocity clauses, not source-available with restrictions — clears the most demanding enterprise legal review tiers. Combined with the architectural detail that V4-Pro is a sparse MoE (1.6T total but only 49B active), the per-token compute footprint sits well below dense flagship comparables. Chinese open-weight share now above 60% of OpenRouter usage is the demand-side confirmation of what V4 enables.

For US-based labs (OpenAI, Anthropic, Google), the strategic question sharpens. Closed-flagship pricing has to justify the gap against MIT-licensed weights at production capability. Anthropic's $30B run-rate proves enterprise customers will still pay; the long-tail enterprise share that won't is what V4 absorbs.

Codersera — best open-source LLM May 2026 → · HuggingFace — best open-source LLMs → · Codersera — open-source LLMs landscape 2026 →