Moving to deepseek-flash-Abliterated on TopxAI
· deepseek, models, api, topxai
Flash moves to a private deployment with a new model ID, fixed $1.20 / $0.15 / $2.40 token prices, image inputs and a 262,144-token output limit.
TopxAI now offers deepseek-flash-Abliterated in place of deepseek-flash. The replacement is a private deployment of a fine-tuned model, with text and image inputs, a 1,048,576-token context window and a maximum of 262,144 output tokens.
The route ID remains deepseek-official for existing keys. That name does not mean this Flash model calls DeepSeek's official API. deepseek-v4-pro continues on its official deployment and keeps its peak/off-peak prices.
Update the model ID and key
Use the exact ID deepseek-flash-Abliterated, including the capital A. The old deepseek-flash ID returns HTTP 403 and has no forwarding alias. Replace it in saved SDK settings and coding clients.
You can keep the same TopxAI base URL and an unrestricted DeepSeek or Auto-route key. If your key has an allowed-model list, edit the list to include the new ID; the upgrade does not broaden a restricted key's permissions. Check any saved output cap against the new 262,144-token maximum.
The model supports /v1/chat/completions, /v1/responses and /v1/messages. Responses supports stateless requests with the conversation context supplied by the client. The model guide includes a request example and explains the output-cap defaults.
Prices no longer change by time of day
Default USD prices per million tokens are:
| Input | Cache read | Output |
|---|---|---|
| $1.20 | $0.15 | $2.40 |
These are the replacement's fixed prices, not the old official Flash tariff. There is no peak/off-peak schedule or long-context premium. Tokens that the deployment reports as cache hits use the cache-read rate. Accepted requests keep the price captured when they were accepted, and historical charges are unchanged. The pricing page and usage log show the applicable rates and token counts.
Review reasoning settings
The model accepts none, low, high and max for reasoning_effort; the default is high. Set the effort explicitly when a client has a saved preference. A private fine-tune can behave differently from the official model, so check the answers and tool behavior your application depends on before switching its traffic.
Because fine-tuning may affect safety alignment, the model may produce biased, offensive or locally unlawful content. Users bear full responsibility for compliance and legal consequences. TopxAI's Acceptable Use Policy applies to this private deployment as it does to GLM-5.3-Abliterated.
Models: deepseek-flash-Abliterated · Models & pricing
Related
- deepseek-flash-Abliterated API pricing
- deepseek-flash-Abliterated
- Create an API key and choose a route
- How requests are billed
- Moving to Claude Sonnet 5.5 on TopxAI
- Moving to GPT-6.1 Sol on TopxAI
- What we fixed in the relay: Claude Code images, fairer billing, fewer failed requests
Other languages: 简体中文