Sign in with ChatGPT, new models, and a simpler model picker

This update connects your ChatGPT account, adds three models, and makes everyday model choice and reading results simpler.

Sign in with ChatGPT

Sign in to Hyperagent with your ChatGPT account on web or iOS. With an eligible ChatGPT subscription, your OpenAI model requests can run on that plan instead of your Hyperagent credits.

The model picker marks models that use your ChatGPT plan. If that plan's allowance runs out, opted-in paid accounts continue automatically on their Hyperagent plan, or you can switch models. Manage the connection in Settings under AI providers. Learn more in Models and compute.

GPT-6.1 Sol, Muse Spark 1.3, and Grok 4.7

Choose GPT-6.1 Sol for demanding coding and agent work with a 950K-token context window. Muse Spark 1.3 handles long documents and codebases, and Grok 4.7 adds xAI's coding and vision model.

Agents and threads set to Sol (Latest) now use GPT-6.1 Sol. Find Muse Spark 1.3 and Grok 4.7 under Manual in the model picker. See Models and compute for limits.

Choose a model by goal

Pick Quality, Balanced, or Low cost and let Hyperagent choose a model for the job. Each profile has its own mark in the composer and thread header, so you can see which goal is active.

Open Manual to pin a specific model, grouped by provider. Schedules and webhooks get a separate Default option that inherits the agent's model. Learn more in Models and compute.

Faster speech generation

Speech and dialogue generation now use Gemini 3.8 Flash-Lite TTS by default. Existing agents and instructions use it automatically, with no changes needed.

Ask for narration or a multi-speaker dialogue as before. See Media generation for voices and limits.

Additional improvements

Transcript folding. Repeated tool calls and document edits fold into one titled group beneath the agent's explanation. Select the group to see each step.
Follow-up previews. Hover over a suggested follow-up to preview the full message before sending it.
Mobile viewer. Tap a document card or file in a conversation on iOS or mobile web to open it in the fullscreen viewer.
Kimi PDF reading. Kimi models read uploaded PDFs as page images, so you can ask about charts, tables, scans, and layout.
Publish button. The sharing dialog's main button reads Publish the first time you share an artifact.
Artifact panning. On a published page opened in its own tab, pan in any direction after zooming in.
Library lookup. Paste an artifact ID, document ID, thread ID, or thread link into Library search to find the matching work.
Clearer tool errors. When a Slack, Google Slides, or MCP tool fails, you see a plain-language explanation and a next step instead of a raw error.