New DeepSeek V4 Pro is Almost as Good as Claude Fable 5 and Super Cheap

DeepSeek has officially published DeepSeek V4 Pro 0813, moving its flagship 1.6-trillion-parameter AI model from preview to general availability.

Key points

• According to DeepSeek's own tests, Terminal-Bench 2.1 rose from 72.1 to 87.9, CyberGym jumped from 52.7 to 83.3, and DeepSWE climbed from 12.8 to 62.7.
• Other documented scores include 61.5 on NL2Repo, 74.1 on Toolathlon-Verified, 31.8 on AutomationBench Public, and 67.2 on DSBench-Hard.
• Independent testing by Artificial Analysis gives V4 Pro 0813 a score of 53 on its Intelligence Index, compared with 60 for Kimi K3.
• During peak hours, V4 Pro now costs $1.32 per million uncached input tokens and $3.96 per million output tokens, compared with the previous $0.435 and $0.87 rates.
• V4 Pro supports a 1-million-token context window and output of up to 384, 000 tokens.

Notably, the model is now available through DeepSeek's app, website, and API under the existing deepseek-v4-pro name. DeepSeek has additionally issued the model weights under the MIT License. DeepSeek Raises Rates by Up to 4x.

DeepSeek notes the new V4 Pro delivers significantly better agent performance than its April preview, despite keeping the same basic model architecture.

These figures should be treated as vendor-reported benchmarks. Independent testing by Artificial Analysis gives V4 Pro 0813 a score of 53 on its Intelligence Index, compared with 60 for Kimi K3. The stronger model additionally comes with significantly higher API costs.

Off-peak pricing drops to $0.66 for input and $1.98 for output. Peak hours run from 01: 00-04: 00 UTC and 06: 00-10: 00 UTC.

Meanwhile, the output cost therefore climbed by roughly 4.5 times during peak hours, marking a significant shift from the aggressive low-cost pricing that supported DeepSeek gain attention in the AI market. Telegram Wants to Offer 1 Billion Users Their Own Domains. 1 Million-Token Context and New Developer Features.

V4 Pro supports a 1-million-token context window and output of up to 384, 000 tokens. It offers both thinking and non-thinking modes, with low, high, and max reasoning-effort settings.

DeepSeek has additionally went on native backing for the OpenAI Responses API format and specific integration with Codex.

Alongside the model, DeepSeek published DeepSeek Harness, an open-source agent framework built around a plugin-based architecture. Developers can replace components such as models, tools, storage, and agent loops. The project is available under the MIT License and remains in developer preview.

Obtain the latest tech news, telecom insights, and product launches wherever you prefer.

Add as a preferredSource on Google Follow on Google News Join WhatsApp.

Add ProPakistani to Preferred Sources and see more of our stories in Google Search and Top Stories.

Technology and Automotive Specialist covering the latest cars, smartphones, AI breakthroughs, and.

Taken together, the developments around new DeepSeek V4 Pro is Almost as Good as Claude Fable 5 point to a situation that is still moving, and the coming days should bring more clarity.