Success rates by model (last 30 days)
Image models
| Model | 30-day success rate |
|---|---|
| Doubao Seedream 5.5 | 99.9% |
| Gemini 3 Pro Image (Pro) | 99.9% |
| Nanobanana2 | 99.9% |
| Qwen Image Edit Spicy | 99.9% |
| Z-Image Spicy Pro | 99.9% |
| gemini-3-pro-image | 99.9% |
| GPT Image 2 | 99.9% |
| Nano Banana 2.1 | 99.9% |
| gemini-3.1-flash-image | 99.8% |
| Z-Image Spicy | 99.8% |
| GPT Image 2.5 Sunburst | 99.7% |
| GPT Image 2.6 Flare | 99.7% |
| Grok Imagine Image 2.0 | 99.7% |
| GPT Image 2.6 Sunburst | 99.6% |
| FLUX.2 Klein 4B | 99.5% |
| Gemini 2.5 Flash Image | 99.5% |
| GPT Image 2.5 Flare | 99.4% |
| Nanobananapro-gemini | 99.2% |
| Qwen Image 3.0 | 99.2% |
| Grok Imagine Image | 99.1% |
| Doubao Seedream 5.0 Flash | 98.8% |
| GPT Image 2 All | 98.5% |
Video models
| Model | 30-day success rate |
|---|---|
| MiniMax H3 Lite | 99.9% |
| Omni Flash (Gemini) | 99.9% |
| Dreamina Seedance 2.0 Mini | 99.6% |
| VEO 3.1 Fast HD | 99.6% |
| Seedance 2.5 | 98.7% |
| Seedance 2.0 Fast | 98.4% |
| Seedance 2.0 Mini | 98.1% |
| Wan 3.0 Video | 98.1% |
Language models
| Model | 30-day success rate |
|---|---|
| DeepSeek V4 Flash | 99.9% |
| DeepSeek V4.1 Flash | 99.9% |
| Gemini 3 Flash Preview | 99.9% |
| Gemini 3.1 Flash Lite | 99.9% |
| GPT-5.6 Luna | 99.9% |
| Claude Haiku 5.5 | 99.9% |
| DeepSeek V4 Pro | 99.9% |
| GPT-6 Astra | 99.9% |
| Claude Haiku 4.5 | 99.8% |
| Claude Sonnet 5 | 99.8% |
| Claude Sonnet 5.5 | 99.8% |
| GPT-5.6 Terra | 99.8% |
| Claude Opus 4.8 | 99.7% |
| GPT-6 Sol | 99.7% |
| GPT-6.1 Sol | 99.7% |
| GPT-6.1 Terra | 99.7% |
| Gemini 3.6 Flash | 99.6% |
| Claude Opus 5.5 | 99.2% |
| Gemini 3.5 Flash | 99.1% |
| GPT-5.6 Sol | 99.0% |
| GPT-5.5 | 98.9% |
| Gemini 3.1 Pro Preview | 98.7% |
| GLM-5.3 | 98.6% |
| GPT-6 Luna | 98.6% |
| Gemini 4 Argon | 98.5% |
| Gemini 3.8 Flash | 98.3% |
| Qwen3.8 Max | 98.3% |
| Claude Opus 5 | 98.1% |
| Claude Sonnet 4.6 | 98.1% |
Audio models
| Model | 30-day success rate |
|---|---|
| ElevenLabs Sound Effects v3 | 99.9% |
| Fish Audio Drama 3 | 99.9% |
| Eleven v3 | 99.9% |
| Eleven v4 Turbo | 99.9% |
| Eleven Multilingual v2 | 99.8% |
| Eleven v4 | 99.8% |
| Minimax Speech 02 Turbo | 99.5% |
| Fish Audio S2.1 Pro | 98.7% |
| Kling Face Recognition | 98.7% |
Embedding models
| Model | 30-day success rate |
|---|---|
| Text Embedding 3 Large | 99.9% |
| Text Embedding 3 Small | 99.9% |
Frequently asked questions
How is the success rate calculated?
It counts finished calls over the last 30 days (2026-10-12 to 2026-11-11): successes divided by successes plus failures. Failures caused on the caller's side, such as content-policy refusals, malformed parameters, insufficient balance and client disconnects, are not counted, because changing the prompt, the parameters or the balance resolves them, and they are never billed. Every other failure counts against us, including errors returned by upstream providers.
Am I charged for failed calls?
Failed calls are not billed. Popular models run on more than one upstream channel, and when the primary one errors, rate-limits or times out, the request moves to a backup channel automatically, with no code change on your side.
Why are some models not in the table?
The table lists models with at least 200 calls in the last 30 days and a success rate of 98% or higher. New or lightly used models do not have enough calls yet and are left out.
Is this a service guarantee?
These are measured historical figures, not a service-level commitment. They change with each refresh, and the period they cover is shown on this page.