A robot is sprinting towards you. Do you want it running on Claude or Grok?
openrouter.ai
June 17, 2026
26 min read
🔥🔥🔥🔥🔥
61/100
Summary
A battle royale simulation with eleven LLMs revealed that Grok 4.1 Fast won 43% of its matches, outperforming competitors. The cheapest model achieved 27 times more cost-effective wins than the most expensive one.
Key Takeaways
- Grok 4.1 Fast won 43% of the matches in a battle royale simulation, achieving a cost per win of $0.97.
- Claude Sonnet 4.6 finished second with only 5 wins, costing $26.78 per win, highlighting a 27x cost difference between the two models.
- GPT 5.4 had the highest kill count with 38 kills but only secured 2 wins, demonstrating a disconnect between kill performance and overall victory.
- Three models, GPT 5.4-mini, DeepSeek 4 Flash, and Kimi K2.6, spent $57 collectively and did not win any games.
Coverage & discussions
Hacker News · openrouter.ai
Community discussions
Hacker News discussion ↗
Historical count: 5 · 5 saved in sample · Historical sample
Analysis ready
Community Sentiment
Positives
- Grok's ability to navigate export control directives could enhance its deployment in real-world scenarios, making it a more viable option for practical applications.
- DeepSeek V4 Flash's performance in coding tasks showcases its efficiency, indicating a strong potential for developers seeking rapid solutions.
- The discussion around the cost efficiency of AI models highlights the importance of balancing performance with financial viability, suggesting a growing awareness of sustainable AI development.
Concerns
- The high cost of using advanced models like Opus 4.7 raises concerns about their financial viability, especially when compared to human labor for similar tasks.
- The phrase 'cost per kill' associated with AI applications in military contexts evokes ethical concerns about the implications of deploying such technologies.
- The suggestion to avoid sprinting robots reflects a fear of their potential dangers, indicating a lack of trust in the safety of advanced AI systems.
Related Articles
Kimi K2.6 just beat Claude, GPT-5.5, and Gemini in a coding challenge
May 3, 2026
AI Built a Nuke and Still Lost
Jun 23, 2026
We made Grok 4.5, GPT-5.5, and Claude build the same apps
Jul 8, 2026
Agent swarms and the new model economics
Jul 20, 2026
"Drawing" the Mona Lisa with GPT-5.6, Claude, Gemini, and Grok
Jul 21, 2026