Weekly · AI model usage · week 45, 2026
deepseek-v4.1-flash + 6 more AI models by real usage · week 45, 2026
What developers actually ran this week, by real usage — deepseek-v4.1-flash, glm-5.3-flash, gpt-6-luna, and more: the most used, the biggest climbers, and the new entrants, drawn out of the noise.
Most Used
DeepSeek V4.1 Flash is a sparse mixture-of-experts model from DeepSeek, and the first built on the company's Causal Encoder-Decoder (CED) architecture. It activates 8B parameters on input and 16B on
1M ctx · 73 tokens this week
View on OpenRouter →What we said about deepseek-v4.1-flash
Developers are leaning on this one for high-volume pipeline work — log parsing, bulk document summarization, and the kind of batch jobs where you're feeding it enormous inputs and want fast, cheap turnaround. That's DeepSeek V4.1 Flash, the sparse mixture-of-experts model that's the first built on DeepSeek's Causal Encoder-Decoder architecture, activating eight billion parameters on input and sixteen on output. The thing that jumps out from the adoption numbers is the pairing of that million-token context window with seventy-three trillion tokens routed through it this week. That's not experimental tinkering — that's teams wiring it into production for jobs where context length and throughput actually matter. When you've got codebases, transcripts, or massive data dumps that need to stay in one window, a cheap Flash variant that holds a million tokens starts to look like infrastructure rather than a toy. Seventy-three trillion tokens tells you a lot of people made exactly that call this week, and kept it running.
GLM-5.3-Flash is a native multimodal model from Z.ai. It is suited for efficient coding and long-horizon agent tasks. Its hybrid sparse and linear attention architecture maintains accurate long-context behavior while
1M ctx · 15 tokens this week
View on OpenRouter →What we said about glm-5.3-flash
Developers are wiring this one into long-running agent loops and multimodal coding pipelines where a request might span hours of back-and-forth rather than a single prompt. That workload is where GLM 5.3 Flash from Z.ai is showing up, and the usage reflects it: fifteen trillion tokens routed through it this week. What stands out is how that number lines up with its design — a million-token context window and a hybrid sparse-and-linear attention setup that holds its footing across really long sessions. Teams are leaning on it to parse screenshots, docs, and code together, then keep that whole picture in frame as an agent works through a multi-step task. The "Flash" label tells you part of the appeal: people are reaching for it when they need to burn through a lot of tokens without the cost of a single interaction spiraling. Fifteen trillion is a serious volume, and it suggests developers aren't just testing it — they're running it in production on the kind of persistent, context-heavy work that used to break smaller windows.
GPT-6 Luna is the fast, cost-efficient model in OpenAI's GPT-6 series, positioned below GPT-6 Sol. It is suited for high-volume and latency-sensitive workloads such as chat, classification, and lightweight agentic
1M ctx · 7 tokens this week
View on OpenRouter →What we said about gpt-6-luna
Developers are wiring this one into chatbots, classification pipelines, and the kind of lightweight agent loops that fire thousands of calls a minute. That's GPT-6 Luna, the fast, cost-efficient member of OpenAI's GPT-6 lineup that sits just below Sol. And the usage here is genuinely eye-opening — seven trillion tokens routed through it this week. That's not a vanity number; that's what happens when teams find something cheap enough and quick enough to put on the hot path without flinching at the bill. With a one-million-token context window, folks are stuffing in long transcripts, big document batches, and sprawling tool histories, then letting it churn through at speed. You can see the pattern in the traffic: this is the model people reach for when latency matters and volume is relentless — support flows, tagging jobs, routing decisions, the unglamorous work that never sleeps. Seven trillion tokens tells you developers aren't just testing Luna. They're running production on it.
Biggest Climber
GPT-6.1 Terra is the mid-tier model in OpenAI's GPT-6.1 series, priced below GPT-6.1 Sol. It is suited for everyday agentic, coding, and writing work, is particularly strong at understanding screenshots
1M ctx · +194 vs last week
View on OpenRouter →What we said about gpt-6.1-terra
Developers are wiring this one into agentic workflows that live halfway between reading the screen and writing the code — think automated QA runs, UI-aware debugging, and docs that update themselves from a screenshot. That's GPT-6.1 Terra, the mid-tier pick in OpenAI's 6.1 lineup, and this week its usage jumped a hundred and ninety-four percent over last week. What's driving that? Partly the price — it sits below Sol, so teams running it at scale aren't flinching at every call. Partly that million-token context window, which lets folks drop entire repos, long transcripts, or a stack of mockups into a single pass without chunking gymnastics. And the screenshot comprehension is clearly finding its people — a lot of the traffic looks like front-end and agent builders feeding it visual state and asking for the next action. When a model nearly triples its runs in a week, that's usually the sign of teams quietly migrating their everyday coding and writing jobs onto something cheaper that still handles their real workload. Worth watching where it lands next week.
pplx-decider-v1.2, on OpenRouter
+167 vs last week
View on OpenRouter →What we said about pplx-decider-v1.2
Developers are routing this one into decision-heavy pipelines — the kind of workflows where you need a model to pick a path, classify an intent, or decide whether to escalate before the heavier lifting kicks in. That's pplx-decider-v1.2, and the usage curve this week is hard to ignore: up a hundred and sixty-seven percent against last week. That's the sort of jump you see when a model stops being a side experiment and starts getting wired into production routing logic. What stands out is where it lands in the stack — not the piece doing the final generation, but the one making the fast call on what happens next. Teams building agent flows and multi-step systems lean on components like this to keep latency down and keep the expensive models focused only on the work that actually needs them. A run-rate climbing this fast usually means people found it reliable enough to trust with that gatekeeping role. Worth keeping an eye on whether this momentum holds next week.
New On The Board
The official release of Hy4 delivers comprehensive and substantial improvements over Hy4 Preview, with particularly strong gains in agentic capabilities and performance across a wide range of long-horizon, real-world productivity
1M ctx · 4 tokens this week
View on OpenRouter →What we said about hy4
Developers are leaning on this one for long-horizon agentic work — the kind of multi-step automation that runs for ages and still has to hold the whole thread together. That's Tencent's Hy4, and the usage signal here is hard to ignore: four trillion tokens moved through it this week. What stands out is where those tokens are going. People are wiring it into real productivity pipelines — agents that plan, act, check their own work, and keep going without a human babysitting every turn. That million-token context window is clearly doing heavy lifting, letting teams feed in sprawling codebases, long document sets, and full task histories in one shot instead of chopping everything into fragments. And since this is the official release following the preview, folks who kicked the tires early are now committing to it for production runs, not just experiments. Four trillion tokens isn't curiosity traffic — that's developers trusting it with workloads that actually matter to their day.
Crouton Alpha is a lightweight model built for scalable coding, agentic, and multimodal workflows. It's ideal for executing agentic sub-tasks, tool use, and document analysis in high-volume, repeated work. Developers
1M ctx · 1 tokens this week
View on OpenRouter →What we said about crouton-alpha
Developers are wiring this one into the unglamorous grind—the repeated sub-tasks, the tool calls, the document passes that run thousands of times a day without anyone watching. Crouton Alpha is showing up as the quiet workhorse behind a lot of agentic pipelines this week, and the numbers tell the story: a full trillion tokens processed. That's not a model people are poking at for fun—that's a model people are leaning on in production. The pull makes sense. With a one-million-token context window, you can hand it entire codebases or long document sets without stitching things together, and it's light enough to fan out across high-volume workflows where cost per call actually matters. When you're executing the same agentic step on repeat, you want something dependable and affordable, not something precious. A trillion tokens in seven days says developers found exactly that fit. It's the kind of adoption that builds quietly—one automated task at a time—until suddenly it's doing a lot of the work.