DeepSeek V4 trails Opus, cuts costs

The real story here is not whether V4 beats Opus on every leaderboard. It is that DeepSeek keeps compressing the quality gap enough to make open-weight models a serious default for teams that care about cost, control, and throughput.

• –DeepSeek’s official release frames V4 around 1M context and agent-focused workflows, which is the practical differentiator, not raw leaderboard bragging rights.
• –Community benchmark chatter puts it below GPT-5.5 and Claude Opus 4.7, but still close enough that many real workflows may not justify the closed-model premium.
• –The economic angle matters most: if you can get near-frontier behavior at materially lower compute cost, the product becomes strategically useful even without a clean SOTA crown.
• –“Open” here mostly means optionality; local running is still expensive enough that most people will use hosted access instead of self-hosting.