TMTPOST — DeepSeek restored full service on its V4 Flash API Tuesday after morning performance degradation triggered by unprecedented traffic volumes. OpenCode, an overseas open-source AI coding-agent platform, reported that DeepSeek V4 Flash faced capacity shortfalls from record access levels and that users might encounter errors while the team worked on an urgent fix. Developers described the official API as nearly unusable for much of the morning, drawing direct comparisons to the overload that hit Kimi K3 immediately after its launch. DeepSeek itself confirmed that the V4 Flash API had suffered performance declines earlier in the day and that the issue had been resolved with service returned to normal. The surge followed the July 31 public-beta release of the V4 Flash-0731 build, which upgraded agent capabilities through post-training while leaving the underlying architecture unchanged; the model carries 284 billion total parameters with 13 billion active, a one-million-token context window, and list pricing of $0.14 per million input tokens on cache miss and $0.28 per million output tokens. Platform data underscored the scale of demand: OpenCode recorded single-day token volumes reaching several trillion on V4 Flash alone in the days after the update, with the model capturing the majority share of observed traffic on the service. The episode highlights persistent infrastructure constraints even as Chinese models close performance gaps on agent and coding benchmarks; concurrency limits on the official endpoint stand at 2,500 for Flash, yet peak loads still produced temporary unavailability. DeepSeek’s rapid restoration leaves the API operational for developers routing coding and agent workloads, though the volume spike demonstrates that capacity planning remains a binding constraint for the lowest-cost frontier endpoints.