1M-context Flash models: MiniMax, DeepSeek, Qwen and GLM

Short answer: prices are close, capabilities differ

LLMTR is a gateway that gives access to language models from different providers through one OpenAI-compatible API. MiniMax M3.1 Flash (minimax/minimax-m3.1-flash), DeepSeek V4.1 Flash (deepseek/deepseek-flash), Qwen3.8-Flash (qwen/qwen3.8-flash) and GLM-5.3-Flash (zai/glm-5.3-flash) are called here on the /v1/chat/completions endpoint at https://llmtr.com/v1 with the same API key from your llmtr.com account. All four have a 1,000,000-token context window, and moving from one to another is, for most requests, a change of the model string.

Input prices range from $0.135 to $0.15 and output prices from $0.45 to $0.60 per 1M tokens. Capabilities decide more of the choice: how long an output you can get in one response, whether you can send video, how you control reasoning, and how strict the JSON output is. This article compares only the facts in the LLMTR catalog; it gives no quality score and does not replace a trial on your own task.

• MiniMax, Model Invocation guide · 2026-10-19

Which 1M-context Flash model should you choose?

If you need video input, the options are MiniMax M3.1 Flash and Qwen3.8-Flash; the DeepSeek V4.1 Flash and GLM-5.3-Flash cards list no video. For work that needs a long output in one response the limits diverge: MiniMax M3.1 Flash 524,288, DeepSeek V4.1 Flash 393,216, and Qwen3.8-Flash and GLM-5.3-Flash 131,072 tokens. If you need JSON that follows a schema, the model whose card lists structured outputs is Qwen3.8-Flash.

If you need to switch reasoning off entirely, Qwen3.8-Flash supports it; on MiniMax M3.1 Flash and GLM-5.3-Flash reasoning cannot be turned off and its depth is chosen with reasoning_effort. If a price that does not change with the time of day matters to you, take DeepSeek V4.1 Flash's weekday peak windows into account. The table puts the four rows side by side with the values in the LLMTR catalog; prices are at standard hours, per 1M tokens, in the order input / cached input / output.

What does the same agent turn cost on each model?

In an agent loop most of the prompt repeats on every turn: the system prompt, the tool definitions and the earlier messages. When 40,000 tokens of a 50,000-token prompt are read from the cache and the answer is 3,000 tokens long, one turn at standard hours costs $0.0039 on MiniMax M3.1 Flash, $0.00342 on DeepSeek V4.1 Flash, $0.00355 on Qwen3.8-Flash and $0.00378 on GLM-5.3-Flash. The same turn in DeepSeek's peak window costs $0.00684.

This calculation assumes the cache hits; the cache is a discount, not a guarantee, and the hit rate can vary between models. The four models also produce different token counts for the same question, which matters because reasoning tokens are billed at the output price. The Python example below sends the same question to all four models and prints the token counts and the cost at standard-hour prices for each. reasoning_effort is sent only to the two models whose cards list the field.

OpenAI Python SDK example that sends the same question to the four models through https://llmtr.com/v1/chat/completions and prints token usage and the cost at standard-hour prices.

import os

from openai import OpenAI

client = OpenAI(base_url="https://llmtr.com/v1", api_key=os.environ["LLMTR_API_KEY"])

# Standard-hour prices per 1M tokens: input, cached input, output.
# DeepSeek doubles every item in its weekday peak windows.
PRICES = {
    "minimax/minimax-m3.1-flash": (0.15, 0.015, 0.60),
    "deepseek/deepseek-flash": (0.15, 0.003, 0.60),
    "qwen/qwen3.8-flash": (0.15, 0.016, 0.47),
    "zai/glm-5.3-flash": (0.135, 0.027, 0.45),
}
# reasoning_effort is sent only to the models that list it.
EXTRA = {
    "minimax/minimax-m3.1-flash": {"reasoning_effort": "low"},
    "zai/glm-5.3-flash": {"reasoning_effort": "low"},
}
messages = [{"role": "user", "content": "List three risks of caching user sessions in Redis."}]

for model, (inp, cached, out) in PRICES.items():
    response = client.chat.completions.create(
        model=model, messages=messages, max_tokens=4096, **EXTRA.get(model, {})
    )
    usage = response.usage
    details = usage.prompt_tokens_details
    hit = (details.cached_tokens or 0) if details else 0
    cost = ((usage.prompt_tokens - hit) * inp + hit * cached + usage.completion_tokens * out) / 1_000_000
    print(model, usage.prompt_tokens, usage.completion_tokens, f"${cost:.6f}")

DeepSeek V4.1 Flash's peak-hour price

At standard hours DeepSeek V4.1 Flash costs the same as MiniMax M3.1 Flash: $0.15 input and $0.60 output. The difference is the hour. In two weekday windows, between 01:00 and 04:00 and between 06:00 and 10:00 UTC, which is 04:00 to 07:00 and 09:00 to 13:00 Turkey time, every item doubles: $0.30 input, $0.006 cached input and $1.20 output. On weekends and Chinese public holidays the standard price applies all day.

Which price a request is billed at is decided by the moment the request reaches LLMTR. If your work is concentrated in Turkish business hours, the second window covers most of the morning; scheduling batch jobs for the afternoon or evening removes the difference. On MiniMax M3.1 Flash, Qwen3.8-Flash and GLM-5.3-Flash the price is the same at every hour of the day.

Video, JSON and reasoning control

The two models whose cards list video input are MiniMax M3.1 Flash and Qwen3.8-Flash. On MiniMax M3.1 Flash, send the video in a video_url part as a base64 data URL. All four models take images in an image_url part. Image and video tokens are counted in prompt_tokens and billed at the input price.

On the JSON side, MiniMax M3.1 Flash and DeepSeek V4.1 Flash return a valid JSON object with type: json_object inside response_format; MiniMax M3.1 Flash does not hold to a strict json_schema. Qwen3.8-Flash lists schema-following output on its card. The GLM-5.3-Flash card has no JSON mode; in a flow that needs JSON, parse and validate the output on your side with this model.

Reasoning control takes four different forms across the four models. On MiniMax M3.1 Flash reasoning cannot be turned off and is set with five-level reasoning_effort; max applies when the field is omitted. On GLM-5.3-Flash reasoning is always on, reasoning_effort accepts low, high and max, and the default is low. On Qwen3.8-Flash reasoning is switched on and off, but no level is listed. On DeepSeek V4.1 Flash reasoning is on by default and the card lists no reasoning_effort level. A reasoning_effort value that a model does not list returns 400, so review this field as well when you change models.

Trying all four with the same key

Open an account on llmtr.com, top up your balance under Dashboard > Billing and create a key under Dashboard > API Keys. Set the base URL in your OpenAI-compatible SDK to https://llmtr.com/v1; the same key works on all four models, and each request is charged to your balance at the catalog price in force when it is processed. Streaming works on all four with stream: true; if you send include_usage: true inside stream_options, usage arrives in the last chunk.

Run the comparison with an evaluation set from your own work: send the same prompts to the four models, put the answers and the token counts in usage side by side, and work out the cost for your own traffic hours. Setting a spending limit on your API key caps the cost during the trial.

Comparing the four Flash models with the same prompt

Send the same prompt to the four models with one LLMTR key, compare token counts and cost, and choose by the capabilities you need.

1. Key. Top up your balance on llmtr.com and create a key under Dashboard > API Keys.
2. Same prompt. Send your own evaluation set with the four model ids to the https://llmtr.com/v1/chat/completions endpoint.
3. Choice. Compare the answers, token counts and cost for your traffic hours, then choose the model id.

Frequently asked questions

How much do the four models cost?

At standard hours, per 1M tokens, input / cached input / output: MiniMax M3.1 Flash $0.15 / $0.015 / $0.60, DeepSeek V4.1 Flash $0.15 / $0.003 / $0.60, Qwen3.8-Flash $0.15 / $0.016 / $0.47, GLM-5.3-Flash $0.135 / $0.027 / $0.45. On DeepSeek every item doubles in the weekday peak hours.

Which of them accept video?

MiniMax M3.1 Flash and Qwen3.8-Flash. The DeepSeek V4.1 Flash and GLM-5.3-Flash cards list no video input.

Which gives the longer output in one response?

MiniMax M3.1 Flash returns at most 524,288 tokens per response, DeepSeek V4.1 Flash 393,216 tokens, and Qwen3.8-Flash and GLM-5.3-Flash 131,072 tokens.

Related posts

• MiniMax M3.1 Flash API: first request, reasoning_effort, price
• MiniMax M3.1 Flash or M3? Price and feature differences
• MiniMax M3.1 Flash vs Haiku 5.5, GPT-6 Luna, Gemini 3.8 Flash
• Which MiniMax model? M3.1 Flash, M3, M2.7 and M2.5

• MiniMax M3.1 Flash model
• DeepSeek V4.1 Flash model
• Qwen3.8-Flash model
• GLM-5.3-Flash model
• DeepSeek peak-hour pricing