GPT-6 Sol is updated to GPT-6.1 Sol, Opus to 5.5, and we also added Sonnet 5.5 and Grok 4.7.

🌙 Meanwhile, GPT-6 Luna remains the best pick for most tasks — it is significantly faster and cheaper than the more powerful models.

When the task is hard and quality matters more than cost, you now have options to switch to. Relative to Luna on the Artificial Analysis thinking index: GPT-6.1 Sol +60%, Opus 5.5 +70%, Sonnet 5.5 +57%, Grok 4.7 +53%.

Extra “thinking” comes at a price: in our tests Sol was 17.5× more expensive than Luna, Grok 48.7×, Sonnet 85.1×, and Opus 165.4×.

They also took longer than Luna to complete the tasks.

Agents can now be configured to react to running automations.

You can choose when the agent should stay out of the conversation: for any automation start, or only for the ones you select.

For example, a keyword automation can send a prepared reply while the agent keeps answering everything else.

Background automations that do not reply to the user can be excluded so they do not interfere with the agent.

✨ Prefer not to configure this by hand? Ask AI Sensei — for example: “Make the agent stop replying when automation <name> starts.”

Funnels now show more than just how many leads moved to the next stage.

Each stage can now be broken down with markers. For example, on a ?Chose a service? stage you can mark which exact service the lead picked.

A marker can be set when an agent moves a lead to a stage, or via an automation step.

As a result, the funnel shows not only how many leads reached a stage, but also what they chose ? which services are more popular and which are less.

And the funnel itself now looks like? a funnel ?? More visuals, less spreadsheet vibe.

More on funnel stage markers: https://senler.io/help/ru/funnels/markers

Use automation messages as the greeting in the website widget.

In the widget you can now pick a Message step from an automation. It is shown to the lead as soon as a new dialog opens — before they start chatting.

The greeting can be dynamic. For example:

• show one message to a new user with no prior dialogs;

• show a different message to a returning user;

• use messages from different automations based on your conditions;

• optionally show such messages not only at the start, but also during the dialog.

More on connecting and configuring the widget: https://senler.io/help/ru/site-actions.

You can also describe the task to the AI sensei on the Senler.io home page — it will explain how to set things up. Even better: connect our MCP server to your development environment so the AI can help embed and configure the widget in your code for your specific scenario.

A new automation step — Typing status — shows the user that the bot or agent is typing while later steps run.

Supported in VK, Telegram, MAX, and Discord. In other channels the step simply adds a delay.

Project settings now include a new section — “Action history”.

You can see what changes team members made in the project. That makes it easier to follow colleagues’ actions and understand who changed what.

New GPT-6 models in Senler.

We updated Luna and Sol: higher reasoning benchmark scores, and token rates at least twice as low as the GPT-5.6 models with the same names. Luna output tokens are 2.4× cheaper. GPT-6 Luna is now the best balance of reasoning, speed, and cost.

The GPT-6 lineup now has three models:

⚡️ Luna — fast and economical: for customer replies and high message volume. Reasoning — 3/5, speed — 5/5.

🧠 Sol — for analysis, hard questions, and multi-step tasks. Reasoning — 4/5, speed — 5/5.

💎 Astra — the strongest reasoning model in the lineup: for the hardest tasks. More expensive than Luna and Sol. Reasoning — 5/5, speed — 4/5.

All three models understand images and work with connected tools.

There is no separate Terra in GPT-6. We automatically replaced:

• Luna 5.6 → Luna 6,

• Sol 5.6 and Terra 5.6 → Sol 6.

GPT-6 Astra was already available and remains the top model. Agent settings were switched to the new models automatically — nothing to change manually.

A new step is available — AI condition.

It uses a new AI type — Jev. It is very fast and extremely cheap, but instead of generating an arbitrary reply it picks the best option from a predefined set. The model also cannot recognize images or audio attachments — if those are present, prefer classic LLM models.

In our test, a single run of this step cost only 0.0021 credits — about 0.00214 ₽.

To use AI condition:

1. add the step to an automation;

2. describe the instruction;

3. specify the possible exit options;

4. choose a confidence threshold — 50% is enough in most cases.

This step is handy for cheap intelligent routing. For example, the AI can judge request complexity: send simple ones to a cheaper model and complex ones to a more powerful model.

AI condition also works well for detecting topic, request type, user intent, and choosing the right automation branch — without running a full AI model every time.

Jev cannot recognize images, audio, or other attachments. If the step logic must consider attachment contents, use classic LLM models that support those formats.

Card auto-payment is now available. In project settings, under Plans and billing, you can link a bank card and turn on automatic plan renewal and credit top-ups when the balance runs low.

Agents and automations can keep running without stopping because a subscription expired or credits ran out.

GigaChat 3.5 Reasoning is now available — Sber's reasoning model.

🧠 Reasoning — 3/5 on our scale. The same rating tier as GPT-5.6 Luna.

⚡️ Speed — roughly on par with Gemini Flash: short answers in our checks took 3–5 seconds. Complex tasks may take longer.

💰 Price — mid-range. Text generation costs almost the same as Gemini Flash; processing incoming text is about 60% more expensive. On uncached token rates, GigaChat is 3–4× more expensive than DeepSeek V4.1 Flash, but substantially cheaper than GPT-6 Astra and Claude Opus 5.

The cost of a specific reply depends on its length and how much reasoning it uses.