DeepSeek and SpaceXAI both released new artificial intelligence agent models on August 12, unveiling the V4 Pro and Grok 4.6 within hours of each other. The launches marked major developments in the AI landscape, with each platform targeting high-performance agent tasks for automation and code generation.
Model pricing and performance
DeepSeek announced that V4 Pro is available at a rate of $0.87 per million output tokens, positioning itself at a fraction of competitors’ costs. In contrast, SpaceXAI listed Grok 4.6 at $6 per million output tokens, with input token pricing set at $2 per million. For those requiring a faster version, prices double, while requests with context lengths above 200,000 tokens increase to $4 for input and $12 for output per million tokens.
Both DeepSeek and SpaceXAI designed their latest models with extended context capabilities and sustained agent work in mind, addressing the growing demand for advanced automation. Grok 4.6 aims to provide efficient multi-step tool invocation, code modification, and deliver completed artifacts. V4 Pro offers an output rate roughly one-seventh that of Grok 4.6 and significantly undercuts flagship models like OpenAI’s GPT-5.6 Sol and Claude Fable 5.
On a technical level, V4 Pro’s architecture is a mixture of experts system with 1.6 trillion total parameters and 49 billion active per token. It was trained on more than 32 trillion tokens. The model can handle a context window of up to one million tokens with a 384,000 token output ceiling. Its open weights have already been downloaded over 1.4 million times on Hugging Face in a single month.
Benchmarks and real-world use
V4 Pro achieved a score of 83.3 on the CyberGym agent benchmark, narrowly surpassing Claude Opus 4.8 and matching or outperforming major competitors. On AutomationBench, V4 Pro scored 31.8, leading its peers. It also posted an 87.9 result on Terminal-Bench 2.1, topping Opus 4.8 but falling just shy of Fable 5 by 0.1 points.
SpaceXAI’s evaluation of Grok 4.6 highlighted its score of 1,753 Elo on GDPVal-AA v2 and strong showing in the Harvey legal benchmark, reaching 15.8%, while GPT-5.6 Sol posted 2.5%. On the Artificial Analysis Intelligence Index, Grok 4.6 tied with GPT-5.6 Sol at 61 points, standing one point behind Fable 5.
DeepSeek V4 Pro delivered notable agent gains over its preview release, with higher scores on DeepSWE (+49.9), CyberGym (+30.6), NL2Repo (+23.0), and Terminal Bench 2.1 (+15.8), according to OpenRouter.
In applied testing, V4 Pro demonstrated lower costs and richer output detail. For example, building a Flappy Bird game used more than 20,000 tokens on V4 Pro at a cost of $0.019, while Grok 4.6 used about 5,000 tokens, costing $0.03. However, V4 Pro did encounter issues such as a reversed pelican animation in one scenario.
DeFi agents and market shifts
The low pricing of V4 Pro has made continuous agent operations in decentralized finance more viable. Running an application that produces 100 million output tokens monthly now costs approximately $87 on V4 Pro, compared to $2,500 on Claude Opus 4.7 or $3,000 on GPT-5.5. DeepSeek reached the $0.87 per million rate after making a 75% price cut permanent in May, but has warned that a “significant increase” in API pricing could come soon.
Considering recent automation trends and the growing dependence on real-world asset tokenization, industry observers note that a broad shift is taking place. Wall Street is progressively transitioning to Web3, with investors able to use platforms like 1stepSwap to own shares of major U.S. companies, gold, and silver directly in their crypto wallets. Tokenized Real-World Assets (RWAs) and automated price-finding mechanisms on these platforms now bypass traditional middlemen entirely, enhancing access and efficiency.
Founder Liang Wenfeng led DeepSeek’s rapid V4 Pro launch, following a four-month preview period. In an internal address, Elon Musk projected SpaceXAI’s AI business would surpass all other SpaceX revenue streams by September and make up 99% of the company’s value within five years.
