Blog

Abliterated models on TopxAI: the November 10 prices and how to call them

2026-11-10 · glm, deepseek, models, pricing, api, topxai

Abliterated models cost more from November 10. The two older ones are still below their launch prices on input and output. Prices and how to call them.

TopxAI sells three abliterated models: GLM-5.3-Abliterated, GLM-5.3-Flash-Abliterated and deepseek-flash-Abliterated. All three run on private deployments, not on the providers' own APIs. Their prices went up on November 10, 2026. This post gives the new prices, sets them beside what each model cost when it first went on sale, and shows how to call the models.

"Abliterated" means the model's built-in refusal alignment has been removed. The Acceptable Use Policy still covers every request, and the compliance notice at the end of this post applies to all three models.

Prices from November 10

USD per million tokens on the main deployment:

The prices are the same at every hour. There is no long-context tier and no cache-write lane: prompt caching is automatic, cached prompt tokens are billed at the cache-read price, and reasoning tokens are billed as output.

Compared with the day before, input and output are up by half on all three models. Cache reads are up by half on GLM-5.3-Abliterated and deepseek-flash-Abliterated, and down from $0.08 to $0.075 on GLM-5.3-Flash-Abliterated. No other model changed price. A request accepted before the change keeps the price captured when it was accepted.

These are modified models on private deployments, and each one costs more than the provider's list price for the unmodified base model. The pricing pages show both prices side by side.

Input and output are still below the launch prices of the two older models

Two of the three models have a price history. Each started higher, dropped sharply on November 8, and has now come part of the way back. Per million tokens, input / cache read / output:

Against the launch price:

• deepseek-flash-Abliterated is 65% lower on input, 30% lower on cache reads and 37.5% lower on output.
• GLM-5.3-Abliterated is 25% lower on input and about 14% lower on output. Cache reads are the exception: $0.45 now, $0.40 at launch.
• GLM-5.3-Flash-Abliterated went on sale only on November 8, so there is no higher earlier price to compare with: its input and output are 50% above that first price. It remains the lowest-priced of the three, at 12% of GLM-5.3-Abliterated's input price and 25% of its output price.

As an example, take a job that reads 1,000,000 input tokens and writes 200,000 output tokens, with no cache hits:

That job now costs 22% less than at launch on GLM-5.3-Abliterated and 57% less on deepseek-flash-Abliterated.

Which model to pick

GLM-5.3-Abliterated is the largest of the three and takes text only. GLM-5.3-Flash-Abliterated is cheaper and faster and reads images. deepseek-flash-Abliterated reads images too, and it is the only one of the three that can switch reasoning off.

Create a key

1. Open API keys and create a key. Under Group, choose GLM for the two GLM models, DeepSeek for deepseek-flash-Abliterated, or Auto route, which reaches all three with one key.
2. If you set Model limits on the key, add each model ID you want to call. A restricted key does not gain a model by itself.
3. Check that Wallet has credit.

The key guide covers quotas, expiry and the other routes.

Make a request

The base URL is https://ai.topxea.com/v1, and the key goes in the Authorization header. Model IDs are case-sensitive: send them exactly as written here, with the capital letters. An ID in the wrong case gets HTTP 403.

Chat Completions with curl:

curl https://ai.topxea.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TOPXAI_API_KEY" \
  -d '{"model":"GLM-5.3-Abliterated","max_tokens":1024,"messages":[{"role":"user","content":"Explain what a race condition is in three sentences."}]}'

The OpenAI Python SDK, with a deeper reasoning setting:

from openai import OpenAI

client = OpenAI(base_url="https://ai.topxea.com/v1", api_key="sk-...")
reply = client.chat.completions.create(
    model="deepseek-flash-Abliterated",
    reasoning_effort="high",
    max_tokens=4096,
    messages=[{"role": "user", "content": "Review this function for bugs: ..."}],
)
print(reply.choices[0].message.content)

An image for GLM-5.3-Flash-Abliterated, sent as a base64 data URL:

curl https://ai.topxea.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $TOPXAI_API_KEY" \
  -d '{"model":"GLM-5.3-Flash-Abliterated","max_tokens":1024,"messages":[{"role":"user","content":[{"type":"text","text":"What is in this picture?"},{"type":"image_url","image_url":{"url":"data:image/png;base64,<BASE64>"}}]}]}'

GLM-5.3-Flash-Abliterated does not fetch an http(s) image URL: the model answers that it cannot see the image, and the request is still billed. deepseek-flash-Abliterated takes both forms. GLM-5.3-Abliterated takes no images; an attachment sent to it is replaced with a short text note and the rest of the message is forwarded.

The two models with a Responses endpoint take POST /v1/responses. Send the whole conversation in input, because the conversation is not stored on the server. All three also answer POST /v1/messages in the Anthropic format, through a conversion. Any client that accepts an OpenAI-compatible base URL and key works the same way; the SDK guide has more examples.

Reasoning

A request that sets no effort runs at low on all three models. Set the depth with reasoning_effort, or reasoning.effort on Responses. low, high and max work on all three. Only deepseek-flash-Abliterated accepts none, which turns reasoning off; on the two GLM models none runs as low, and the reasoning tokens are still spent.

The relay fits the other OpenAI values onto these: minimal runs as low, medium as high and xhigh as max. Reasoning tokens are counted in the completion tokens and billed at the output price, so a higher effort costs more per answer.

Limits to plan for

• n is not supported. One choice comes back and is billed once.
• logprobs are not returned.
• Input and output share the context window, and one answer can be at most 131,072 tokens.
• Tools, parallel tool calls, streaming with usage, temperature, top_p and JSON output work as in the OpenAI format. The model guides list the details: GLM models, deepseek-flash-Abliterated.

Backup deployment

GLM-5.3-Abliterated and deepseek-flash-Abliterated each have a standby deployment. If the main deployment fails before it answers, a request may move to the standby automatically. That is rare, and the request is then billed at the backup price: $4.00 / $0.40 / $7.00 for GLM-5.3-Abliterated and $1.20 / $0.15 / $2.40 for deepseek-flash-Abliterated. These backup prices did not change on November 10. GLM-5.3-Flash-Abliterated has no standby: if its deployment fails, the request fails and nothing is charged.

The usage log shows what each request was charged, with the token counts and the rates applied, and marks a request the standby served as Backup deployment.

Compliance notice

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.

Removing the refusal alignment does not remove the Acceptable Use Policy. If you are unsure whether a use case is allowed, ask through the help button at the bottom right or write to support@topxea.com before building on it.

Models: GLM-5.3-Abliterated, GLM-5.3-Flash-Abliterated, deepseek-flash-Abliterated · Models & pricing

Related

• GLM-5.3-Abliterated API pricing
• GLM-5.3-Flash-Abliterated API pricing
• deepseek-flash-Abliterated API pricing
• GLM-5.3-Abliterated and GLM-5.3-Flash-Abliterated
• deepseek-flash-Abliterated and deepseek-flash-fast
• Create an API key and choose a route
• How requests are billed
• Moving to deepseek-flash-Abliterated on TopxAI
• Moving to Claude Fable 5.2 on TopxAI
• GPT-6.1 Terra on TopxAI

Other languages: 简体中文