DeepSeek launched the official version of its V4 Pro AI model, marked 0813, on August 13, offering significantly upgraded agent capabilities compared to previous versions [1, 2, 3, 4]. The launch was sudden and occurred without formal blog posts or announcements [2, 3, 4].
DeepSeek presented V4 Pro as a major step forward, with benchmark scores approaching or slightly surpassing those of Anthropic’s Claude Fable 5 model in some AI agent tests. The V4 Pro scored 87.9 in Terminal Bench 2.1, very close to Claude Fable 5’s 88 [2, 3, 4]. DeepSeek highlighted the much lower price, with token output costs at $0.87 per million tokens, about 1/45 to 1/60 the price of Fable 5’s $50 per million tokens [1, 2, 3, 4]. According to an implied DeepSeek official, "V4 Pro's benchmark scores approach or exceed those of Anthropic's Claude Fable 5, with much lower pricing" [3].
The model supports advanced features including a one million token context window, JSON output, tool calls, and API compatibility [4]. The price for V4 Pro usage rose roughly fivefold compared to previous DeepSeek models, though still significantly below competing high-end AI models [1, 2, 3, 4].
However, less than 24 hours after release, DeepSeek quietly removed V4 Pro from public availability. Users reported poor real-world performance with low-level logic errors, graphical glitches, and overall unsatisfactory results[c3]. Some users suspected that the public deployment may have mistakenly routed requests to the older V4 Flash model rather than the new Pro version[c3]. DeepSeek has not issued any official statement explaining the removal or addressing performance concerns as of August 13[c3].
The rollout coincided with xAI’s release of the Grok 4.6 AI model, both emphasizing agent-based capabilities and competitive pricing [4].
DeepSeek plans further price increases for V4 Pro API usage, as indicated on its official pricing pages [4].
The controversy continues as DeepSeek remains silent on whether the withdrawal was due to deployment errors or genuine flaws. Meanwhile, users await clarification and a potential re-release once issues are resolved [3].
