{"ok":true,"source":"tensorfeed.ai","upstream":"Source: OpenRouter (openrouter.ai/rankings), as of 2026-09-28, usage data through 2026-09-27. Licensed under CC BY 4.0. sharePct is each row's share of the ranked rows' tokens (93.18T), not of all OpenRouter traffic, because the public page no longer states a platform total.","window":"rolling 7 days (complete UTC days 2026-09-21 to 2026-09-27)","lastUpdated":"2026-09-28","count":10,"rankings":[{"rank":1,"model":"DeepSeek V4.1 Flash","provider":"DeepSeek","openrouterId":"deepseek/deepseek-v4.1-flash","tokensB7d":19600,"trend":"up","sharePct":21,"notes":"Released 2026-09-10 as DeepSeek's first Causal Encoder-Decoder model (552B total, 8B active on input, 16B on output), MIT weights. Took the top spot in its first full week, up 24%.","url":"https://api-docs.deepseek.com/news/news260910"},{"rank":2,"model":"GLM 5.3 Flash","provider":"Z.ai","openrouterId":"z-ai/glm-5.3-flash","tokensB7d":16300,"trend":"up","sharePct":17.4,"notes":"First natively multimodal GLM-5 model (320B total, 18B active, MIT). Up 16% week over week, second by tokens behind DeepSeek V4.1 Flash.","url":"https://z.ai/blog/glm-5.3-flash"},{"rank":3,"model":"Space Bunny Alpha","provider":"Stealth","openrouterId":"stealth/space-bunny-alpha","tokensB7d":13900,"trend":"new","sharePct":14.9,"notes":"Anonymous stealth model OpenRouter listed on 2026-09-23, free for the preview, 1M context with text, image and video input. Reached #3 on five days of traffic.","url":"https://openrouter.ai/stealth/space-bunny-alpha"},{"rank":4,"model":"Hy4 preview","provider":"Tencent","openrouterId":"tencent/hy4-preview","tokensB7d":9640,"trend":"down","sharePct":10.3,"notes":"Tencent MoE preview released August 2026: 770B total, 49B active, Apache-2.0 weights. Down 23% week over week as newer flash tier models took share.","url":"https://huggingface.co/tencent/Hy4-preview"},{"rank":5,"model":"GPT-5.6 Luna","provider":"OpenAI","openrouterId":"openai/gpt-5.6-luna","tokensB7d":8530,"trend":"down","sharePct":9.1,"notes":"Most cost-efficient GPT-5.6 tier and the highest-volume closed model in the window. Down 12% week over week, with GPT-6 Luna arriving 2026-09-22.","url":"https://openai.com/index/gpt-5-6/"},{"rank":6,"model":"DeepSeek V4 Flash 0731","provider":"DeepSeek","openrouterId":"deepseek/deepseek-v4-flash-0731","tokensB7d":7820,"trend":"down","sharePct":8.3,"notes":"Re-post-trained July 31 revision of V4 Flash with MIT weights. Down 17% week over week as DeepSeek traffic moves to V4.1 Flash.","url":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"},{"rank":7,"model":"Nemotron 3 Ultra (free)","provider":"NVIDIA","openrouterId":"nvidia/nemotron-3-ultra-550b-a55b:free","tokensB7d":5670,"trend":"up","sharePct":6,"notes":"Free OpenRouter variant of NVIDIA's open 550B MoE (55B active, OpenMDW-1.1 license). Up 26% week over week.","url":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16"},{"rank":8,"model":"MiMo-V2.6-Flash","provider":"Xiaomi","openrouterId":"xiaomi/mimo-v2.6-flash","tokensB7d":5510,"trend":"new","sharePct":5.9,"notes":"Xiaomi omnimodal MoE released 2026-09-21 with MIT weights: 309B total, 15B active, 1M context, text, image, audio and video input. Entered at #8, taking the MiMo-V2.5 slot.","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL"},{"rank":9,"model":"DeepSeek V4 Flash","provider":"DeepSeek","openrouterId":"deepseek/deepseek-v4-flash","tokensB7d":3350,"trend":"down","sharePct":3.5,"notes":"Original April 2026 checkpoint (OpenRouter label: DeepSeek V4 Flash 0423), 284B total with 13B active, MIT weights. Down 11% week over week.","url":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"},{"rank":10,"model":"GPT-6 Luna","provider":"OpenAI","openrouterId":"openai/gpt-6-luna","tokensB7d":2860,"trend":"new","sharePct":3,"notes":"OpenAI's cheapest GPT-6 tier, launched 2026-09-22 at $0.10 input / $0.50 output per 1M with a 1.05M context window. Entered the top 10 on five days of traffic.","url":"https://openai.com/index/introducing-gpt-6-sol-and-luna/"}]}