XNews

DeepSeek V4 Pro Official Release and Price Hike

DeepSeek V4 Pro Official Release and Price Hike

DeepSeek released the DeepSeek-V4-Pro-0813 model and announced API price changes on August 13, 2026. An earlier 0831 V4 Pro reference caused confusion; API and model documents showed 0813 while the update log did not mention 0813, and a website banner for an official version update was removed. No official announcement accompanied the initial appearance, and some reports called it a silent release with no official X post and no PR campaign. A formal 0813 update announcement later appeared, confirming the release and pricing change. Cline testing placed Terminal-Bench 2.1 at 87.9, second behind Claude Fable 5 by 0.1 points. Comparisons stated V4 Pro performance exceeded Opus 4.8 and was close to Fable 5; two metrics surpassed Fable and multiple surpassed Opus 4.8. Cost comparisons before the increase cited V4 Pro as 57 times cheaper than Fable 5 or 100 times cheaper than Fable. API pricing shifts to peak and off-peak from August 17 00:00; cache-hit input prices rise 12x and output prices rise 4.5x. Peak Pro output pricing is more than 4 times the previous level; off-peak Pro pricing is more than double. Artificial Analysis showed official V4 Pro only one point above V4 Flash, with nearly identical scores. The same day also included the DeepSeek Harness announcement with plan mode, sub-Agents, permission approval, MCP, and main-loop plugins.

Key insights

Conflicting release signals for V4 Pro 0813

The DeepSeek V4 Pro 0813 model first appeared with documentation and pricing referring to 0813, while the update log did not mention 0813. The official website banner for an update to the DeepSeek V4 Pro official version was removed. No official WeChat article or public announcement was available at first. A formal 0813 update announcement arrived later, after initial confusion over whether the model had been updated or pulled. Early reports also mentioned an 0831 version reference that did not match other pages.

Benchmark performance against Claude Fable 5 and Opus 4.8

Cline team testing placed DeepSeek-V4-Pro-0813 at 87.9 on Terminal-Bench 2.1, second behind Claude Fable 5 by 0.1 points. Reports stated that V4 Pro performance exceeded Opus 4.8 and was close to Fable 5, with two metrics surpassing Fable and multiple metrics surpassing Opus 4.8. The model was called strong and inexpensive.

API pricing restructured with peak and off-peak rates

DeepSeek API pricing changed to peak and off-peak rates starting August 17 at 00:00. Cache-hit input prices increased by 12 times and output prices by 4.5 times. Peak pricing for Pro model output was more than 4 times the previous price, and off-peak Pro pricing was more than double the previous price. Separate charts and summaries were circulated to explain the changes for V4 Pro and V4 Flash.

Cost advantage before the price increase

Before the announced increase, DeepSeek-V4-Pro-0813 was described as 57 times cheaper than Claude Fable 5 in one test-based comparison and 100 times cheaper than Fable in another. These price-performance comparisons were cited as reasons for praise of the release.

Artificial Analysis score nearly equal to V4 Flash

Artificial Analysis results showed the DeepSeek V4 Pro official version scored only one point above V4 Flash, with nearly identical scores. Early observers said this score did not align with expectations and added to uncertainty about the official release.

DeepSeek Harness announced alongside V4 Pro

On the same day, DeepSeek announced DeepSeek Harness. The system converts planning mode, sub-Agents, permission approval, MCP integrations, and the agent main loop into removable and replaceable plugins. This was listed as one of three major announcements from DeepSeek.

Source posts