Kimi K3 vs DeepSeek V4 Pro vs GLM-5.2: Open Trillion-Scale MoE Models Compared on Benchmarks, License, and Serving Cost

Three Chinese labs now maintain the highest of the open-weight leaderboard. Moonshot AI’s Kimi K3, DeepSeek V4 Pro, and Zhipu AI’s GLM-5.2 are all sparse Mixture-of-Experts (MoE) fashions with million-token context home windows. Each targets long-horizon coding and agent workloads. This article compares them on three axes an AI staff really decides on: measured functionality, license phrases, and serving value.

‘Trillion-parameter’ suits Kimi K3 (2.8T) and DeepSeek V4 Pro (1.6T). GLM-5.2 is 744B complete, so it’s the smallest of the three by complete parameters. It earns its place as a result of it led the open-weight area earlier than K3 shipped.

Which mannequin for which job

For lowest value per token at robust coding high quality, DeepSeek V4 Pro is the clear choose. Its weights are downloadable, its license is clear, and its output value undercuts each rivals.

For the best measured functionality, Kimi K3 leads, however at 5x to 17x the output value and no downloadable weights till July 27. GLM-5.2 sits between them: cheaper than K3, quicker than each rivals, self-hostable as we speak, and extra succesful than its dimension suggests.

If you might be planning to decide on based mostly on verification depth and license readability favor DeepSeek and GLM now. Buyers chasing peak benchmark scores look forward to K3 weights or pay the API premium.

Key Takeaways

• Kimi K3 leads the Artificial Analysis Intelligence Index (~57, #3 general) however stays API-only till July 27.
• DeepSeek V4 Pro is the fee chief: ~$0.04 per process and ~1.15M output tokens per greenback at listing charges.
• GLM-5.2 (744B) is the smallest but quickest (~168 t/s) and self-hostable as we speak beneath MIT.
• All three ship 1M-token context; solely DeepSeek and GLM have open weights accessible now.

The put up Kimi K3 vs DeepSeek V4 Pro vs GLM-5.2: Open Trillion-Scale MoE Models Compared on Benchmarks, License, and Serving Cost appeared first on MarkTechPost.