AI Pulse · AI Trends Pulse

DeepSeek just released V4 Flash with output pricing around $0.28 per million tokens. Compare that to roughly $25 per million for Claude Opus 4.8, and you’re looking at a 99% discount. The company also claims it now leads the Arena coding leaderboard, though that assertion comes from social media commentary rather than direct Arena verification, so treat it as developing rather than settled fact.

The pricing gap matters because coding tasks can burn through tokens fast. If you’re running automated code reviews, generating documentation, or prototyping features at scale, a two-order-of-magnitude price difference changes the economics completely. What was prohibitively expensive at $25 per million tokens becomes background noise at $0.28. You can suddenly afford to throw the model at problems you wouldn’t have touched before.

The leaderboard claim is harder to pin down. Arena rankings shift constantly, and unverified X posts aren’t the same as published benchmarks. That said, even if DeepSeek sits somewhere in the top tier rather than strictly first, the combination of competitive performance and radically lower cost is what shifts the calculation for operators. You don’t need the absolute best model if a near-peer costs 1% as much and lets you run ten times the volume.

This is exactly the kind of trade-off analysis we build into the Omni Command Centre, where you can route tasks to the most cost-effective model for each job rather than locking into a single vendor at a single price point. The model landscape changes weekly. Your infrastructure shouldn’t assume it’s static.

Watch for independent benchmarks over the next few days. If the coding performance holds up under scrutiny, DeepSeek just made a lot of enterprise AI budgets look very different. You can read the original report for the full technical detail.