What happened
In mid-July 2026, DeepSeek moved its V4 model family out of preview, standardizing a one-million-token context window across the lineup and improving agent, math, and coding performance. Alongside it, DeepSeek introduced peak and off-peak API pricing for the first time, charging double the off-peak rate during weekday busy hours. The V4 line spans a larger Pro model and a smaller, cheaper Flash model.
The context
Time-of-day pricing is a new lever for managing demand, and a sign of how strained inference capacity has become.
Why this moves the Board: Service change · Materiality 3 · T3 Reported → minor input for DeepSeek.