Auto mode: the best model for every task, automatically
Today we’re launching Auto model routing, available to all your Tasklet agents. It picks the best model for each task and message, so you only pay for top-model power when the task actually needs it. We’re also launching Ultrafast, a new speed setting that works with Auto mode and select models.
New models come out all the time, and no single one is best at everything. Our customers kept asking which model to use for what, and every new release started a debate on our own team too. It’s nearly impossible to keep track of all the changes, so you either overpay for simple tasks or get second-rate results on the work that matters most.
Auto · Gemini 3.8 FlashAuto · Gemini 3.8 FlashGemini 4 Argon$$$$$Gemini 3.8 Flash$$$$$Gemini 3.5 Flash Lite$$$$$
GLM 5.3 Flash$$$$$How it works
Auto mode is a model router built into Tasklet. Every message you send goes to the best model for the job, judged on quality first and then cost. The difference adds up fast: GPT-6 Astra costs 100x more per token than GPT-6 Luna, so a quick lookup doesn’t need to run on a top model. Auto mode decides message by message, so a single thread can move between models as the work changes. It also weighs the cost of switching models mid-task, so it only switches when the change is worth it.
Our team built the routing logic from our own testing and data on which models give the best results for the price. Jev, a fast and inexpensive model, applies that logic to every message, so routing never slows your agent down.
It keeps getting better
Your agents always benefit from the best models available. As models get better and cheaper, your agents get better results at lower cost automatically.
Tasklet gives your agents access to models from OpenAI, Anthropic, Google, xAI, Moonshot, and Z.ai. Today, Auto mode chooses leading models from OpenAI and Anthropic for most tasks, like GPT-6 Luna for quick lookups, GPT-6.1 Sol for everyday work, Claude Opus 5.5 for design work or polished reports, and GPT-6 Astra for complex and detailed analysis.
We keep tuning the router as new models launch and new capabilities arrive, so you can focus on getting work done instead of tracking who’s winning the AI race this week.
Need to move quick? Go Ultrafast
Auto mode picks the model, you pick the speed. When you’re waiting on an answer and every second counts, open the model picker, go to Speed, and choose Ultrafast. It works with Auto mode, so you still get the best model for each message, just faster. For now, Ultrafast runs on GPT-6.1 Sol and GPT-6 Astra, and other models run at Fast or Standard speed.
When responses feel instant, you stay in the flow instead of switching to something else while your agent works. Ultrafast uses about 6x the credits, so Auto mode never turns it on for you. You decide when speed is worth it.
Try Auto mode
New agents use Auto mode by default. You can switch an existing conversation by clicking the model name in the message box and choosing Auto from the model picker. You can pick a specific model the same way whenever you want.
If you’ve set a specific default model before, here’s how you can switch it to Auto:
- Go to Settings → General (under My account).
- Set Default model to Auto.
With Auto mode, every message gets the best model for the job at the lowest price that does it well, and it keeps improving without you changing a thing.