In progress
A wave of new models, and price cuts
RouteMux has recently added a batch of new text, image and video models, and several models now cost less. Price cuts come first, then each new model's price and discount by category. These are the prices at the time of this post; each model page has the current price.
Price cuts
| Model | Was | Now | Change |
|---|---|---|---|
| DeepSeek V4.1 Flash | $0.30 / $1.20 | $0.15 / $0.60 | -50% (50% off official) |
| DeepSeek V4 Flash → V4.1 Flash | $0.44 / $1.32 | $0.15 / $0.60 | input -66%, output -55% |
| DeepSeek V4 Flash Vision Exp → V4.1 Flash | $0.44 / $1.32 | $0.15 / $0.60 | input -66%, output -55% |
| MiniMax H3 · 768p | $0.08 / s | $0.04 / s | -50% |
| MiniMax H3 · 2K | $0.13 / s | $0.065 / s | -50% |
| Grok 4.7 | $0.40 / $1.20 | $0.30 / $0.90 | -25% (85% off official) |
| Grok 4.6 | $0.40 / $1.20 | $0.30 / $0.90 | -25% (85% off official) |
- Text model prices are USD per million tokens, input / output; Grok prices are for prompts under 200K tokens ($0.60 / $1.80 above that).
- DeepSeek V4.1 Flash is now 50% off the official price. DeepSeek has folded V4 Flash into V4.1 Flash, and we follow: calls that use the old names keep working without code changes, are routed to V4.1 Flash and are billed at the V4.1 Flash price. V4.1 Flash accepts image input as well.
Text models
USD per million tokens, input / output.
| Model | Price | Official price | Discount |
|---|---|---|---|
| Gemini 4 Argon | $1.00 / $5.00 | $2.00 / $10.00 | 50% off |
| Qwen3.8-Max | $1.40 / $4.20 | $2.00 / $6.00 | 30% off |
| Qwen3.8-Flash | $0.105 / $0.329 | $0.15 / $0.47 | 30% off |
| MiniMax M3.1 Flash | $0.045 / $0.18 | $0.15 / $0.60 | 70% off |
| Muse Spark 1.3 Contributor | $0.05 / $0.10 | $0.10 / $0.20 | 50% off |
| DeepSeek V4.1 Flash | $0.15 / $0.60 | $0.30 / $1.20 | 50% off |
- All six have a 1M-token context. Qwen3.8, MiniMax M3.1 Flash and DeepSeek V4.1 Flash support OpenAI Chat, Responses and Anthropic Messages; Gemini 4 Argon supports OpenAI Chat and the Gemini native format; Muse Spark supports OpenAI Chat and Responses.
- Requests to and responses from Muse Spark 1.3 Contributor may be used by Meta to train future models. That is why it costs less; do not send sensitive data to it.
- DeepSeek V4.1 Flash is DeepSeek's newest Flash model and accepts image input; on October 17 DeepSeek upgraded it in place to the 1017 version, same model id.
- Gemini 4 Argon, Qwen3.8, Muse Spark and DeepSeek V4.1 Flash require a top-up.
Image models
Billed per image, all through the OpenAI Images API (/v1/images/generations and /v1/images/edits).
| Model | Price | Official price | Discount |
|---|---|---|---|
| Nano Banana 2.1 · 1K | $0.025 | $0.0336 | 26% off |
| Nano Banana 2.1 · 2K | $0.04 | $0.0504 | 21% off |
| Nano Banana 2.1 · 4K | $0.08 | $0.113 | 29% off |
| Nano Banana Pro · 1K / 2K | $0.10 | $0.134 | 25% off |
| Nano Banana Pro · 4K | $0.18 | $0.24 | 25% off |
| Grok Imagine Image | $0.006 | $0.02 | 70% off |
| Grok Imagine Image 2.0 | $0.012 – $0.024 | $0.04 – $0.08 | 70% off |
| GPT Image 2.5 Sunburst · 2K / 4K | $0.08 / $0.12 | — | — |
| GPT Image 2.5 Flare · 2K / 4K | $0.08 / $0.12 | — | — |
- Grok Imagine Image 2.0 can edit images (up to 3 source images) and is priced by quality × resolution, up to 2K; each source image costs $0.003.
- GPT Image 2.5 Sunburst and Flare both add 2K / 4K output, for text-to-image and image-to-image; the 1K price is unchanged. OpenAI does not publish an official per-image price.
- Nano Banana Pro is back, now through the OpenAI Images API; call it as gemini-3-pro-image or gemini-nano-banana-pro. The old Gemini-native image endpoint is no longer offered.
- Nano Banana 2.1 / Pro and the Grok Imagine image models require a top-up.
Video models
Billed per second, through the /v1/videos API.
| Model | Price | Official price | Discount |
|---|---|---|---|
| Grok Imagine Video 1.5 · 480p | $0.04 / s | $0.08 / s | 50% off |
| Grok Imagine Video 1.5 · 720p | $0.07 / s | $0.14 / s | 50% off |
| Grok Imagine Video 1.5 · 1080p | $0.125 / s | $0.25 / s | 50% off |
| MiniMax H3 · 768p | $0.04 / s | $0.08 / s | 50% off |
| MiniMax H3 · 2K | $0.065 / s | $0.13 / s | 50% off |
| MiniMax H3 Highspeed · 768p | $0.03 / s | $0.06 / s | 50% off |
| MiniMax H3 Lite · 480p | $0.0125 / s | $0.025 / s | 50% off |
| MiniMax H3 Lite · 768p | $0.02 / s | $0.04 / s | 50% off |
| MiniMax H3 Max · 480p | $0.025 / s | $0.05 / s | 50% off |
| MiniMax H3 Max · 768p | $0.04 / s | $0.08 / s | 50% off |
- Grok Imagine Video 1.5: text-to-video, with an optional first frame, last frame and up to 4 keyframes inside the clip, plus up to 14 reference images and 3 preset voices; with sound, 1–15 seconds. Up to 720p when reference images or voices are used.
- Grok Imagine Video 1.5: each input image (first frame, last frame, keyframe or reference image) costs $0.005, against $0.01 official; preset voices are free.
- All four MiniMax H3 models do text-to-video, image-to-video from a first frame or first and last frames, and reference generation with up to 9 reference images, 3 reference videos and 3 reference audio clips. Reference video is billed per input second; reference audio is free.
- MiniMax H3: 4–15 seconds at 768p or 2K. Reference video costs the same as output; the first 5 reference images are free, then $0.02 each ($0.04 official).
- MiniMax H3 Highspeed (new): balances quality and speed, 4–15 seconds, 768p only. Reference video costs the same as output; the first 5 reference images are free, then $0.015 each ($0.03 official).
- MiniMax H3 Lite (new): the lightweight version, 4–15 seconds at 480p or 768p. Reference video costs the same as output; the first 5 reference images are free, then $0.01 each ($0.02 official).
- MiniMax H3 Max (new): post-trained from H3, 5–15 seconds at 480p or 768p. Reference video has its own rate: $0.02765 / s at 480p and $0.0715 / s at 768p ($0.0553 / $0.143 official); the first 2 reference images are free, then $0.037 each ($0.074 official).
- All of these models require a top-up.
Older models merged
These older models have been folded into newer ones: DeepSeek follows the official provider, and Grok 4.5 moves to the newer 4.7, which now costs less for input and output. Calls that use the old names keep working without code changes; they are routed to the newer model and billed at its price:
| Old model | Now routed to | Was | Now |
|---|---|---|---|
| DeepSeek V4 Flash | DeepSeek V4.1 Flash | $0.44 / $1.32 | $0.15 / $0.60 |
| DeepSeek V4 Flash Vision Exp | DeepSeek V4.1 Flash | $0.44 / $1.32 | $0.15 / $0.60 |
| Grok 4.5 | Grok 4.7 | $0.40 / $1.20 | $0.30 / $0.90 |
- After the move, cached input on Grok 4.5 calls goes from $0.06 to $0.075 per 1M tokens, slightly more.
Retired
- GPT-5.5: calls point you to GPT-6.1 Sol.
- The two older Gemini image models (Nano Banana, Nano Banana 2): use Nano Banana 2.1 instead.
- Grok Imagine Image Quality: use Grok Imagine Image 2.0 instead.