Mercury 3 vs Claude Haiku 5.5, GPT-6 Luna, Gemini 3.8 Flash

Short answer: Mercury 3 for text-only agents, the others for images and documents

LLMTR is a gateway that gives access to language models from different providers through one OpenAI-compatible API. Mercury 3 (inception/mercury-3), Claude Haiku 5.5 (anthropic/claude-haiku-5.5), GPT-6 Luna (openai/gpt-6-luna) and Gemini 3.8 Flash (google/gemini-3.8-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 context window of about 1 million tokens, and moving from one to another is, for most requests, a change of the model string.

Mercury 3 is Inception Labs' diffusion-based language model; Inception states that it generates more than 1,000 tokens per second. It accepts text only, and its price depends on the date: $0.10 input and $0.35 output per 1M tokens through the end of 25 October 2026 (TRT), then $0.20 and $0.70 from 26 October 2026. Claude Haiku 5.5 and GPT-6 Luna are $0.10 input and $0.50 output on short prompts; during the introductory period Mercury 3 matches them on input and is below them on output, and from 26 October both cost less on short prompts. What makes you pick Gemini 3.8 Flash is input types: it is the only one of the four that accepts audio and video.

• Inception, Introducing Mercury 3 · 2026-10-19
• Inception, model comparison and pricing · 2026-10-19
• Anthropic, Claude Haiku 5.5 overview · 2026-10-19
• OpenAI, GPT-6 Luna model page · 2026-10-19

Mercury 3, Claude Haiku 5.5, GPT-6 Luna or Gemini 3.8 Flash?

Three questions decide it: which input types you send, how long your prompts are, and whether the answer or the prompt is the longer part. For a text-only agent, search or voice assistant flow that is sensitive to latency, Mercury 3 is a candidate. If you send images, the options are Claude Haiku 5.5, GPT-6 Luna and Gemini 3.8 Flash; for PDF, GPT-6 Luna and Gemini 3.8 Flash; for audio and video, Gemini 3.8 Flash.

On long prompts the thresholds come into play: Claude Haiku 5.5 bills the whole request at a higher price once the prompt passes 100,000 tokens, and GPT-6 Luna once it passes 272,000 tokens; the prices of Mercury 3 and Gemini 3.8 Flash do not change with prompt length. The table shows the four rows with the values in the LLMTR catalog; prices are per 1M tokens, in the order input / cached input / output.

• Google, Gemini 3.8 Flash model page · 2026-10-19

What does the same request cost on each model?

Without a cache hit, and assuming the same token counts on all four models, a 20,000-token prompt with a 2,000-token answer costs $0.0027 on Mercury 3 at the introductory price, $0.0054 from 26 October 2026, $0.003 on Claude Haiku 5.5 and GPT-6 Luna, and $0.0225 on Gemini 3.8 Flash. When the answer is longer than the prompt the order changes: a 2,000-token prompt with a 20,000-token answer costs $0.0072 on Mercury 3 at the introductory price, $0.0144 after 26 October, $0.0102 on Haiku 5.5 and Luna, and $0.0765 on Gemini 3.8 Flash.

With a 300,000-token prompt and a 2,000-token answer, Haiku 5.5 and GPT-6 Luna move to their higher tier: Mercury 3 costs $0.0307 at the introductory price and $0.0614 at the standard rate, GPT-6 Luna $0.0615, Haiku 5.5 $0.155 and Gemini 3.8 Flash $0.2325. Read this calculation with one limit in mind: the four models use different tokenizers, and reasoning tokens are billed at the output price. The Python example below sends a report to the four models and prints each response's time, its token counts and the cost by prompt threshold. reasoning_effort: low is accepted by all four models.

OpenAI Python SDK example that sends the same report to the four models through https://llmtr.com/v1/chat/completions and prints response time, token usage and the cost by prompt threshold.

import os
import time
from datetime import datetime, timezone

from openai import OpenAI

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

# Per 1M tokens: (input, output), the prompt threshold, (input, output) above it.
# A prompt over the threshold bills the whole request at the second pair.
# Mercury 3: introductory price through 25 October 2026 (TRT), standard rate from 26 October.
# Gemini 3.8 Flash: introductory price through 31 December 2026.
INTRO_ENDS = datetime(2026, 10, 25, 21, 0, tzinfo=timezone.utc)
MERCURY_3 = (0.10, 0.35) if datetime.now(timezone.utc) < INTRO_ENDS else (0.20, 0.70)
PRICES = {
    "inception/mercury-3": (MERCURY_3, None, None),
    "anthropic/claude-haiku-5.5": ((0.10, 0.50), 100_000, (0.50, 2.50)),
    "openai/gpt-6-luna": ((0.10, 0.50), 272_000, (0.20, 0.75)),
    "google/gemini-3.8-flash": ((0.75, 3.75), None, None),
}
with open("report.txt", encoding="utf-8") as f:
    document = f.read()
messages = [{"role": "user", "content": "List the five key findings in this report:\n\n" + document}]

for model, (base, threshold, above) in PRICES.items():
    started = time.perf_counter()
    response = client.chat.completions.create(
        model=model, messages=messages, max_tokens=4096, reasoning_effort="low"
    )
    seconds = time.perf_counter() - started
    usage = response.usage
    inp, out = above if threshold and usage.prompt_tokens > threshold else base
    cost = (usage.prompt_tokens * inp + usage.completion_tokens * out) / 1_000_000
    print(model, f"{seconds:.1f}s", usage.prompt_tokens, usage.completion_tokens, f"${cost:.6f}")

Which one when?

Each point rests on a value in the LLMTR catalog; compare answer quality and response time on your own examples.

• A text-only agent or a voice assistant turn that is sensitive to latency: inception/mercury-3. Per Inception's statement it generates more than 1,000 tokens per second; time the responses on your own requests.
• Image input: anthropic/claude-haiku-5.5 or openai/gpt-6-luna; PDF: openai/gpt-6-luna; audio and video: google/gemini-3.8-flash. Mercury 3 accepts text only.
• After 26 October 2026, work with short prompts and long answers: anthropic/claude-haiku-5.5 or openai/gpt-6-luna. Output is $0.50 per 1M tokens, against $0.70 on Mercury 3.
• Prompts over 272,000 tokens: inception/mercury-3 or openai/gpt-6-luna. After 26 October a 300,000-token prompt with a 2,000-token answer costs $0.0614 and $0.0615 on the two, and $0.155 on Haiku 5.5.
• More than 128,000 tokens of output in one response: inception/mercury-3, with a 131,072-token limit.

Reasoning, sampling parameters and the introductory price

Reasoning control differs across the four models. Mercury 3 accepts minimal, low, medium and high, applies low when the field is omitted, and has no none; send minimal to cut latency. Claude Haiku 5.5 and GPT-6 Luna accept six levels, and none turns reasoning off. Gemini 3.8 Flash accepts low, medium and high and defaults to medium. A level that a model does not list returns 400; review this field as well when you change models.

Claude Haiku 5.5 and GPT-6 Luna do not support temperature and top_p. If you send these fields the request is not rejected; the fields are removed and appear in the llmtr_dropped_parameters list of the response. On Mercury 3, repeated prompt prefixes can be served from the cache; cached input is a tenth of the input price.

Mercury 3's introductory price applies through the end of 25 October 2026 (TRT); from 26 October 2026 the standard rate applies on its own, and you do not need to change anything in your requests. Gemini 3.8 Flash's introductory price runs through 31 December 2026, and from 1 January 2027 every item doubles.

• Google, Gemini API pricing · 2026-10-19

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, you do not need a separate account with each of the four vendors, and each request is charged to your balance at the catalog price in force when it is processed.

Make the choice with an evaluation set from your own work: look at the length distribution of your prompts and answers, work out the share of requests that pass the 100,000 and 272,000-token thresholds, and put the four models' answers side by side with their response times and the token counts in usage. Setting a spending limit on your API key caps the cost during the trial.

Comparing the four models with the same prompt

Send the same prompt to the four models with one LLMTR key, compare response time, token counts and cost, and choose by what 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, response times, token counts and cost, then choose the model id.

Frequently asked questions

How much do the four models cost?

Per 1M tokens, input / cached input / output: Mercury 3 through 25 October 2026 $0.10 / $0.01 / $0.35, from 26 October 2026 $0.20 / $0.02 / $0.70; Claude Haiku 5.5 up to 100,000 tokens $0.10 / $0.01 / $0.50, above that $0.50 / $0.05 / $2.50; GPT-6 Luna up to 272,000 tokens $0.10 / $0.01 / $0.50, above that $0.20 / $0.02 / $0.75; Gemini 3.8 Flash through 31 December 2026 $0.75 / $0.075 / $3.75.

Does Mercury 3 accept images or PDF?

No, Mercury 3 accepts text only. For images use Claude Haiku 5.5, GPT-6 Luna or Gemini 3.8 Flash; for PDF, GPT-6 Luna or Gemini 3.8 Flash; for audio and video, Gemini 3.8 Flash.

When does Mercury 3's introductory price end?

At the end of 25 October 2026 (TRT). From 26 October 2026, $0.20 input, $0.02 cached input and $0.70 output per 1M tokens apply; the switch happens on its own.

Related posts

• Mercury 3 vs Mercury 2.5 Preview vs Mercury 2
• MiniMax M3.1 Flash vs Haiku 5.5, GPT-6 Luna, Gemini 3.8 Flash

• Mercury 3 model
• Claude Haiku 5.5 model
• GPT-6 Luna model
• Gemini 3.8 Flash model
• GPT-6 Sol and Luna documentation
• Google Gemini documentation