Guozhen AIGlobal AI field notes and model intelligence

Realtime AI News

MiniMax token usage jumps 690% week over week, enters global top eight

A third-party weekly tally shows MiniMax's models surging 690% in token consumption week over week and climbing to eighth on the global large-model token usage ranking. The same edition shows Xiaomi's MiMo-V2.6-Flash dropping off the list, a reshuffling that points to how quickly real developer traffic moves.

Published

A weekly leaderboard tracking global token consumption by large language models has shifted: MiniMax's models posted a 690% week-over-week jump in token usage, rising to eighth place and becoming the most conspicuous gain on this edition of the ranking. The same list shows Xiaomi's MiMo-V2.6-Flash dropping out.

The figures come from a third-party weekly aggregation of model API traffic, counting the tokens actually consumed when a model is called. It measures neither benchmark quality nor user sentiment; it records how many requests developers and applications route to which models. That is why rank changes are read as a direct window into real-world usage.

A nearly sevenfold increase is a violent move at weekly granularity. Short-term spikes in token volume usually trace to one of three things: migration after a new version ships, pricing or free-quota changes on the developer side, or a single application suddenly scaling up. Nothing in the available information says which of these drove this jump, so the safer reading is a strong demand-side signal rather than a settled conclusion.

Xiaomi's MiMo-V2.6-Flash falling off the list is the other half of the same dataset. The ranking gives relative position only, not reasons: a model leaving the top tier may mean rivals pulled traffic away, or simply that it stayed flat while others sprinted. Read together, the edition looks like a reshuffling at the top rather than a verdict on either vendor.

Token volume sits closer to real usage intensity than download counts or poll results, and a rank that holds for several weeks usually means production applications have wired a model into their pipelines. A single-week surge, by contrast, can be the byproduct of integration testing or a short promotion. Whether MiniMax's jump holds depends on whether it stays in the top ten over the coming weeks.

It is worth noting that the ranking's methodology, the platforms it covers, and how it deduplicates traffic are not disclosed in the available information, which makes any single edition a trend clue rather than a conclusion. The things to watch: whether MiniMax converts this traffic spike into durable share, whether Xiaomi returns to the list, and whether more vendors start trading places on the same board.

Why it matters

The swing shows the market increasingly watches real token consumption rather than launch-day benchmarks. But weekly data is noisy, and staying in the top ten across consecutive editions is the more reliable test.

MiniMax大模型Token调用
Back to realtime news

Nearby Updates

All

11/02, 12:00

Muse AI agent launches in Singapore with Grab, Shopee and Singapore Airlines connectors

Muse, a personal AI agent built to handle everyday tasks, is now available in Singapore on iOS, Android and the web at muse.ai. Its first local connector partners span shopping, food delivery, travel and lifestyle, including Shopee, Grab, Singapore Airlines, Agoda, ShopBack, Klook, Trip.com and Lazada.

11/02, 12:32

DeepSeek open-sources dsh-superagent, a proactive coding agent prototype

DeepSeek's Harness team has open-sourced dsh-superagent, a research prototype that takes development requests from GitHub or Feishu and works on them in separate agent sessions. Released under the MIT license, it turns authorized requests into work items and returns results such as pull requests to the original conversation.

11/02, 12:47

Claude Opus 5.5 Tops SemiAnalysis GDPx Benchmark, Gemini 4 Carbon Second

OfficeChai reports that SemiAnalysis' GDPx benchmark places Claude Opus 5.5 first and Gemini 4 Carbon second. The report discloses the ordering but not the underlying scores, leaving the leaderboard a directional signal rather than a full capability profile.

11/02, 12:54

Software Cost Alliance Releases Cost Evaluation Standard for AI Application Projects

The Software Cost Alliance has released a document titled Cost Evaluation for Artificial Intelligence Application Development Projects in Systems and Software Engineering, relayed by Sina's mobile news channel. The publication pushes AI project costing toward a shared industry vocabulary instead of vendor-by-vendor quotes.