cryptopolitan.com
13 August 2026 12:54, UTC
On August 12, DeepSeek released V4 Pro at $0.87 per million output tokens. The same day, SpaceXAI released Grok 4.6 at $6 per million output tokens.
The two releases came within hours of each other.
Founder Liang Wenfeng shipped the production V4 Pro build early yesterday. The preview ran about four months.
4.8 based on 40K reviews in the App Store and Google Play
SpaceXAI, the AI arm led by Elon Musk, released Grok 4.6 the same day.
Both companies targeted their models for long-horizon agent work. The model continues to invoke tools, modify code, examine its results, and deliver a complete artifact.
V4 Pro’s output rate is about one-seventh of Grok 4.6, one-thirty-fifth of OpenAI’s GPT-5.6 Sol, and one-fifty-seventh of Claude Fable 5.
SpaceXAI lists Grok 4.6 at $2 per million input tokens and $6 per million output. A fast version is double the price, and long context requests over 200,000 tokens are $4 and $12.
DeepSeek beats that, listing input at $0.435 per million on a cache miss and as low as $0.003625 on a cache hit.
V4 Pro is a mixture of experts system with 1.6 trillion total parameters, 49 billion active per token, trained on more than 32 trillion tokens.
The context window is a million tokens with a 384,000 token output ceiling. Open weights were downloaded over 1.4 million times on Hugging Face in a month.
V4 Pro wins on security and automation. It scored 83.3 on the CyberGym agent benchmark, against 83.1 for Fable 5 and 78.3 for Claude Opus 4.8. On AutomationBench V4 Pro hit 31.8, ahead of both.
V4 Pro scored 87.9 on Terminal-Bench 2.1. The score is higher than Opus 4.8’s 85.0 but falls short of Fable 5 by 0.1 points.
SpaceXAI’s tests indicate that Grok 4.6 tops GDPVal-AA v2 with 1,753 Elo, up from Grok 4.5’s 1,526, and scored 15.8% on the Harvey legal benchmark, where GPT-5.6 Sol scored 2.5%.
Grok 4.6 ties GPT-5.6 Sol at 61 on the Artificial Analysis Intelligence Index, one point behind Fable 5.
In a Flappy Bird game build, V4 Pro used 20,000+ tokens for $0.019 and delivered richer detail. Grok 4.6 used about 5,000 tokens but cost $0.03.
V4 Pro also failed with a pelican animation. The bird appeared to be moving backwards.
DeepSeek V4 Pro 0813 is live on OpenRouter.@deepseek_ai reports large agent gains over V4 Pro Preview: DeepSWE 62.7 (+49.9), CyberGym 83.3 (+30.6), NL2Repo 61.5 (+23.0), and Terminal Bench 2.1 87.9 (+15.8)
More providers coming online soon
Use it now: https://t.co/1PYoEFbvyK
— OpenRouter (@OpenRouter) August 12, 2026
Agents executing multi-step trades or watching DeFi positions run all the time.
Cryptopolitan has previously reported that an app that produces 100 million output tokens per month costs ~$87 on V4 Pro compared to ~$2,500 on Claude Opus 4.7 or $3,000 on GPT-5.5.
V4 Pro’s rates can make money for strategies that lose money on Western AI models.
DeepSeek reached the $0.87 rate in May, when it made a 75% cut permanent, dropping output from $3.48, as Cryptopolitan reported at the time.
DeepSeek has stated that a “significant increase” in overall API pricing is coming, so the current floor may not hold.
In a staff address on August 12, Musk said AI revenue would eclipse all other SpaceX lines combined in September and represent 99% of the company’s value in five years.
