Model Retirements

Retired and deprecated models

Every model TokenRa has stopped serving, or will stop serving, with the date and the model to move to.

• Retired: no longer served. Requests that use the model ID fail with HTTP 503 and model_not_found.
• Deprecating: still served, with a retirement date announced. Move to a replacement before that date.

Dates are in UTC. A retired model is no longer served; a deprecating model still answers requests until its date.

MiMo V2.5 (mimo-v2.5): retired

TokenRa retired mimo-v2.5 on October 30, 2026. It is no longer returned by GET /v1/models, and requests that use its ID fail with HTTP 503 and model_not_found.

MiMo V2.5 was Xiaomi's multimodal reasoning model. On TokenRa it took text and image input, returned its reasoning in reasoning_content, and worked in the Chat Completions, Responses and Anthropic Messages formats.

Model ID

mimo-v2.5

Provider

Xiaomi

Status

Retired

Retired on

October 30, 2026

API response now

HTTP 503, model_not_found

Replacement

Qwen3.8 Flash (qwen3.8-flash)

Archived docs

MiMo V2.5 API docs

What to use instead of MiMo V2.5

Qwen3.8 Flash is the closest match. Like MiMo V2.5 it takes image input, returns its reasoning in reasoning_content, enforces json_schema output and honours max_completion_tokens, in all three API formats. DeepSeek V4.1 Flash matches on the same points. GLM-5.3 Flash differs: it ignores max_completion_tokens, so cap its output with max_tokens.

Behaviour on TokenRa as recorded on each model's docs page, which lists the live requests behind it.

Switch the model ID

In most requests the only change is the model field. The replacements also work in the OpenAI Responses and Anthropic Messages formats (/v1/responses, /v1/messages), so nothing else in your integration has to change. Re-run your own prompts before you ship the switch.

cURL

curl https://tokenra.io/v1/chat/completions \
  -H "Authorization: Bearer $TOKENRA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-flash",
    "messages": [{"role": "user", "content": "Explain in two sentences what an API gateway does."}]
  }'

Python (OpenAI SDK)

from openai import OpenAI

client = OpenAI(base_url="https://tokenra.io/v1", api_key="YOUR_TOKENRA_API_KEY")

response = client.chat.completions.create(
    model="qwen3.8-flash",  # was "mimo-v2.5"
    messages=[{"role": "user", "content": "Explain in two sentences what an API gateway does."}],
)
print(response.choices[0].message.content)

What a retired model ID returns

When a request names a retired model, TokenRa answers every API format the same way: HTTP 503 with the error code model_not_found. We checked /v1/chat/completions, /v1/responses and /v1/messages on October 31, 2026.

HTTP/2 503

{
  "error": {
    "code": "model_not_found",
    "message": "分组 default 下模型 mimo-v2.5 无可用渠道（distributor） (request id: ...)",
    "type": "new_api_error"
  }
}

The message text comes back in Chinese. It reads: no available channel for model mimo-v2.5 under group default. TokenRa returns the same error for a misspelled ID or a model your key's group cannot use, so compare the ID with the table above.

Rejected requests are not billed, because they never reach a provider. In our check on October 31, 2026, they created no usage record and cost nothing.

Model retirement FAQ

Continue exploring