---
title: "Grok 4.6, PixVerse V6 and C1, and New Seedance 2.5 Modes Are Now Live on Yotta AI Gateway"
slug: new-on-yotta-ai-gateway-grok-4-6-pixverse-seedance-2-5-modes-2026
description: "Grok 4.6 joins the Gateway’s text catalog, PixVerse V6 and C1 add two video models from $0.017/s, and Seedance 2.5 picks up reference-to-video, video edit, and video extension endpoints. What shipped, what it costs, and the first request for each."
author: "Yotta Labs"
date: 2026-09-18
categories: ["News"]
canonical: https://www.yottalabs.ai/post/new-on-yotta-ai-gateway-grok-4-6-pixverse-seedance-2-5-modes-2026
---

# Grok 4.6, PixVerse V6 and C1, and New Seedance 2.5 Modes Are Now Live on Yotta AI Gateway

![](https://cdn.sanity.io/images/wy75wyma/production/274034b1c810a8a23f693e771871c17d58fcaa45-1200x627.png)

*Since the Seedance 2.5 launch: one frontier text model, two video models, and three new ways to drive Seedance 2.5. Here’s the catalog as of September 18, the prices, and how to call each one.*

The Yotta AI Gateway catalog is now 49 models behind one API key. Since [Seedance 2.5 and 2.0 went live](https://www.yottalabs.ai/post/seedance-2-5-and-2-0-yotta-ai-gateway-video-api) at the end of August, the additions have been: Grok 4.6 on the text side, PixVerse V6 and PixVerse C1 on the video side, and three new endpoints for Seedance 2.5 (reference-to-video, video edit, and video extension), with Seedance 2.0 picking up the edit and extension endpoints too. Wan 3.0 also serves reference-to-video, which puts seven models in that category.

Everything below runs through the same account, key, and bill as the rest of the catalog. Prices are the Gateway’s list prices as of publication; the [pricing page](https://www.yottalabs.ai/pricing) is the live version.

## What’s new at a glance

<!-- unsupported block: table -->

## Grok 4.6: the first Grok on the Gateway

Grok 4.6 is the model string grok-4.6, listed under publisher SpaceXAI. It’s a reasoning model that always thinks before it answers; there’s no way to turn reasoning off, but you can set how hard it works with reasoning_effort (low, medium, high, or xhigh; high is the default). Messages accept text and image_url content, the system prompt caps at 10,000 characters, and max_completion_tokens goes up to 500,000, which only counts visible output, not reasoning tokens.

Pricing has two tiers by prompt size:

<!-- unsupported block: table -->

Keeping prompts under 200K tokens matters twice: it halves the input rate and the output rate at the same time. For comparison inside the catalog, Kimi K3 is $3 in / $15 out and GLM 5.3 is $1.40 / $4.40, so Grok 4.6 lands between them on input and under Kimi on output.

Chat completions use the OpenAI-style endpoint with a bearer token:

```bash
curl -X POST "https://gateway.yottalabs.ai/api/maas/chat/completions" \
  -H "Authorization: Bearer $YOTTA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "grok-4.6",
    "messages": [
      {"role": "user", "content": "Summarize the tradeoffs of tensor parallelism vs pipeline parallelism in three sentences."}
    ],
    "reasoning_effort": "medium",
    "max_completion_tokens": 1024,
    "stream": false
  }'

```

Streaming works the usual way ("stream": true, SSE, data: [DONE] at the end), and reasoning comes back as a separate reasoning_content field in the stream so you can show or hide it.

## PixVerse V6 and C1: the cheap tier of video

PixVerse’s two current models cover text-to-video, image-to-video, and reference-to-video, six endpoints in total (pixverse-v6-t2v, pixverse-v6-i2v, pixverse-v6-r2v, and the pixverse-c1- equivalents). They are the lowest-priced video models on the Gateway by a wide margin, billed per second and tiered by resolution and whether you generate audio:

<!-- unsupported block: table -->

Clips run 1 to 15 seconds. V6 is the newer model and the only one with multi_shot; on reference-to-video it accepts up to 10 images and up to 2 videos (15 seconds total, and video references double the price of the request). C1 is the character-consistency model and takes up to 7 image references. A 10 second 720p clip with audio is $0.41 on V6, against $2.50 on Seedance 2.5 text-to-video ($0.25/s), so PixVerse is where drafts, social cuts, and volume work should go. The full breakdown, including parameters and sample calls for all six endpoints, is in [PixVerse V6 vs C1](https://www.yottalabs.ai/post/pixverse-v6-vs-c1-differences-api-2026).

## Seedance 2.5: three new endpoints

At launch, Seedance 2.5 and 2.0 were text-to-video and image-to-video. Both now have video-edit and video-extension endpoints, and 2.5 also has reference-to-video. All three take a reference_items array and a prompt that refers to those assets by number (@Image1, @Video1, @Audio1).

**Reference-to-video** (seedance-2.5-r2v, endpoint /reference-to-video/generations) is the omni-reference mode: up to 30 images, up to 10 videos (2 to 30 seconds each, 30 seconds total), and up to 10 audio clips (30 seconds total) in a single request. Aspect ratio defaults to adaptive and adds 21:9 to the usual set; resolution is 480p, 720p, or 1080p; duration is 4 to 30 seconds or -1 for smart duration. Pricing depends on whether a video is among the references: $0.12 / $0.25 / $0.59 per second without a video reference at 480p / 720p / 1080p, and $0.08 / $0.15 / $0.36 with one.

**Video edit** (seedance-2.5-video-edit, endpoint /video-edit/generations) requires at least one video in reference_items and rewrites it from the prompt while keeping what you tell it to keep: camera motion, composition, color, the original audio bed. Videos are 4 to 30 seconds. $0.08 / $0.15 / $0.36 per second.

**Video extension** (seedance-2.5-video-extension, endpoint /video-extension/generations) continues a clip you reference in the prompt, picking up from its last frame. Same price as edit. Seedance 2.0 has both edit and extension too, from $0.05 per second.

A reference-to-video request looks like this:

```bash
curl -X POST "https://gateway.yottalabs.ai/api/maas/reference-to-video/generations" \
  -H "X-API-KEY: $YOTTA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seedance-2.5-r2v",
    "parameters": {
      "prompt": "A 20 second product reveal. Open on @Image1 as the first frame, slow push in, then cut to the product on a turntable in the style of @Image2. Score it with @Audio1.",
      "reference_items": [
        {"url": "https://example.com/hero.png", "type": "image"},
        {"url": "https://example.com/style.png", "type": "image"},
        {"url": "https://example.com/bed.mp3", "type": "audio"}
      ],
      "aspect_ratio": "16:9",
      "resolution": "720p",
      "duration": 20,
      "generate_audio": true
    }
  }'

```

Poll GET /reference-to-video/generations/{request_id} until status is completed and read output_url. Edit and extension poll their own paths the same way. The submit-and-poll pattern, with a Python client, is in the [Seedance API guide](https://www.yottalabs.ai/post/how-to-use-seedance-api-generate-video-2026).

## Wan 3.0 reference-to-video

Wan 3.0’s reference endpoint (wan3.0-r2v) is the other omni-reference option: up to 10 images, 5 videos (15 seconds total), and 5 audio clips, numbered the same way, at Wan 3.0’s flat $0.035 / $0.07 / $0.14 per second for 480p / 720p / 1080p, audio included. Clips go to 30 seconds. It’s the cheapest way to run a reference-driven 30 second generation on the Gateway; Seedance 2.5 is the higher-quality, higher-priced version of the same job. [Wan 3.0 vs Wan 2.7](https://www.yottalabs.ai/post/wan-3-0-vs-wan-2-7-differences-pricing-2026) has the parameters and a sample call.

## The video catalog now

Seven models serve reference-to-video: Seedance 2.5 and 2.0, Wan 3.0 and 2.7, PixVerse V6 and C1, and HappyHorse-1.0. Ten serve text-to-video and eight image-to-video. Four have a video edit endpoint, including Seedance 2.5, Seedance 2.0, and Wan 2.7, and two have video extension (Seedance 2.5 and 2.0).

The practical routing, cheapest to most capable: PixVerse V6 for drafts and volume from $0.017/s, Wan 3.0 for 30 second clips and document-driven prompts at $0.035/s, Seedance 2.0 for 4K finals ($0.79/s at 4K), Seedance 2.5 for long-form storytelling and the heaviest reference work ($0.12 to $0.59/s on text-to-video). Because it’s all one key and one bill, switching a pipeline from one to another is a one-line change to model. [Seedance 2.5 vs Seedance 2.0](https://www.yottalabs.ai/post/seedance-2-5-vs-seedance-2-0-differences-which-to-use-2026) covers the choice between the two ByteDance models.

## Get started

Sign in to the [console](https://console.yottalabs.ai/), create an API key, and every model above is in the model library with a playground and a code sample. If you’re new to the Gateway, the [AI Gateway page](https://www.yottalabs.ai/ai-gateway) explains routing, keys, and billing.

## Frequently asked questions

**What is Grok 4.6 on the Yotta AI Gateway?** Grok 4.6 is available as grok-4.6 through the Gateway’s chat completions endpoint at $2 per million input tokens and $6 per million output tokens for prompts up to 200K tokens, double that above 200K, with cached input at $0.50. Reasoning is always on; reasoning_effort controls how much.

**Can I turn off reasoning on Grok 4.6?** No. The model always reasons before answering. Set reasoning_effort to low for the fastest, cheapest responses.

**How much do PixVerse V6 and C1 cost?** V6 starts at $0.017 per second at 360p without audio and tops out at $0.078 per second at 1080p with audio. C1 runs $0.020 to $0.082. A 10 second 720p clip with audio is about $0.41 on V6.

**What does Seedance 2.5 reference-to-video accept?** Up to 30 images, 10 videos, and 10 audio clips per request, referenced in the prompt as @Image1, @Video1, @Audio1. Output is 4 to 30 seconds at 480p, 720p, or 1080p.

**What’s the difference between video edit and video extension?** Edit takes a clip and changes it according to the prompt, keeping whatever you tell it to preserve. Extension takes a clip and generates what happens next, continuing from the last frame. Both require at least one video reference and both are $0.08 to $0.36 per second on Seedance 2.5.

**Which models support reference-to-video?** Seven: Seedance 2.5, Seedance 2.0, Wan 3.0, Wan 2.7, PixVerse V6, PixVerse C1, and HappyHorse-1.0.

**Does my existing Gateway key work for all of this?** Yes. Text models use Authorization: Bearer, video endpoints use X-API-KEY, and both take the same key from the same account.

## Bottom line

The Gateway now has a frontier reasoning model from a fourth lab, a video tier that starts at under two cents a second, and reference, edit, and extension modes on its flagship video model. If you’re already on the Gateway, nothing about your setup changes; the new models show up in the library and in your usage dashboard next to everything else. If you’re not, the [AI Gateway page](https://www.yottalabs.ai/ai-gateway) is where to start, and [Yotta’s pricing page](https://www.yottalabs.ai/pricing) has the full per-model rates.