DeepSeek R1 Outshines Kimi k1.5

This was the moment consumer AI in China stopped rewarding distribution alone and started rewarding the model that felt most exciting to use right now. Moonshot had already spent heavily to make Kimi mainstream, but DeepSeek turned a model launch into a user acquisition event by shipping R1 as a live web, app, and API product with open weights and low token pricing on January 20, 2025, which pulled attention away from Kimi k1.5 even though both launched the same day.1 2 3 4

• DeepSeek gave users and developers three reasons to switch at once. R1 was not just a paper release, it went live immediately in chat, app, and API, it exposed reasoning mode through a simple model endpoint, and it was MIT licensed, so startups could use and adapt it without asking permission.1 2
• Moonshot’s k1.5 looked strong on research positioning, but DeepSeek won the public narrative. Kimi k1.5 surfaced mainly through a GitHub release and benchmark claims, while R1 spread as a product people could try instantly and as a base model developers could distill and deploy, which made it feel bigger than a benchmark matchup.3 4 1
• The fallout showed up in app demand. Research on China’s AI app market tied DeepSeek’s early 2025 surge to the popularity of V3 and R1, while Kimi’s downloads later fell sharply as general chatbots lost momentum. That helps explain why Moonshot pivoted from consumer ads toward enterprise distribution and open weights.5 6 7

From here, model labs that want durable demand will package launches so they spread through both consumers and developers on day one. Moonshot’s later shift toward open models and revenue sharing through Fireworks AI, Together AI, and AWS Bedrock points to the new playbook, win production workloads, not just chatbot buzz.6 7 8