DeepSeek V4: 1M tokens of context and 1.6T parameters — what changes?

Friends, an update from the AI front: DeepSeek previewed V4 Flash and V4 Pro.

Brief:
- Both are mixture-of-experts with up to 1M-token context windows, suited for large codebases and documents.
- V4 Pro: 1.6T parameters (49B active). V4 Flash: 284B (13B active).
- Company reports improved reasoning and coding performance, though knowledge lag is ~3–6 months.
- Models are cheaper than rivals; allegations of "distillation" and IP concerns have emerged.

Why it matters: larger context and lower cost speed LLM adoption, but legal and quality risks persist.

How do you assess market readiness to work with these models?

#AI #LLM #MachineLearning #DeepSeek