Accessibility Adjustments

Use these optional tools to adjust reading and display preferences. These tools cannot resolve every accessibility barrier. Please contact the website owner if you need assistance.

  • Text adjustments
  • Content scaling 100%
  • Font size 100%
  • Line height 100%
  • Letter spacing 100%
  • Colour adjustments
  • Orientation adjustments

HeyGen Decision 1 reaches OpenRouter, with separate video API limits

Listen to this article

HeyGen Decision 1 gives developers structured answers rather than generated prose. OpenRouter dates its standard model listing November 13, 2026, with text and image inputs. HeyGen’s own API also documents video evaluation, but that does not establish video support on OpenRouter. Choose the access route before designing a workflow around its limits.

OpenRouter prices standard and Turbo separately

OpenRouter lists $0.14 per million input tokens and zero output-token charges for Decision 1, with a 131,072-token context window. Its Decision 1 Turbo listing, also dated November 13, quotes $0.09 per million input tokens and zero for output, with 262,144 tokens of context. Turbo remains labeled preview.

Both listings identify HeyGen as the serving provider and list text and images, not video. Neither is currently available through OpenRouter’s EU in-region routing. These are provider listings and access conditions, not evidence of a new November 15 launch or an independent reliability test.

At the displayed rates, 100,000 billed input tokens would cost $0.014 on standard or $0.009 on Turbo. That arithmetic excludes other applicable charges and is not a measured cost per completed workflow. Do not apply OpenRouter’s prices automatically to HeyGen’s direct billing.

Direct access is not the video generation API

HeyGen’s Decision documentation specifies POST /v3/models/decisions, returning results synchronously without polling. Requests accept 1 to 400 named questions. Access requires an API key with assets:write; trial keys and OAuth are unsupported. The direct Turbo preview is not yet open to every API key.

Three answer types cover different jobs: noul estimates yes/no probability, choice selects a supplied category, and score rates an ordered rubric. The text guide separates evidence in state from the question instructions. A JSON array is shared material, not a batch of independent requests. Questions cannot consume one another’s answers; dependent steps need another request.

Video means sampled frames, not automatic transcription

The direct video guide allows one video alongside four images on standard Decision 1. Turbo’s documented preview limits rise to ten videos and 100 images. Video is read through sampled frames. To evaluate speech, supply its transcript as text instead of assuming the soundtrack is understood.

For a product-review workflow, that distinction separates checking visible packaging from checking a spoken claim. Test them separately. A frame-based check can also miss a brief visual event between samples; a positive result should not be treated as proof that every moment was inspected.

Confidence still needs validation

HeyGen warns that confidence is derived from the answer distribution, not calibrated as the chance of correctness. Its 16 MiB request-body limit is separate from context capacity. Reported input usage may count shared material repeatedly across model reads. The default direct limit is 50 requests per second per user within a workspace, and retries have no supported idempotency key.

Start with labeled examples and reversible suggestions, such as assigning draft asset categories for review. Track false approvals and unnecessary review referrals, not just average confidence. Compare the same evidence and acceptance rules when testing alternatives. Our Gemini Decision 1 guide covers another structured-decision API, not an accuracy ranking against HeyGen.

ByteForward has checked the published documentation and listings but has not run inference or measured accuracy, latency or workload cost.

Jordan Reid
Jordan Reid

Jordan Reid is focused on AI tools, agents, developer products, and the way technology changes everyday work. Jordan approaches a launch from the user’s side of the screen. What can it actually help someone finish? The voice is practical, conversational, and skeptical of products that turn a simple job into five new settings. Coverage follows coding assistants, creative software, browser agents, and the workflows around them, with attention to pricing, permissions, setup, and the human work that remains.

Leave a Reply

Your email address will not be published. Required fields are marked *

Gravatar profile