GPT-6.1 Sol and Sol Pro API: measured against GPT-6 Sol

GPT-6.1 Sol is slower and wordier than GPT-6 Sol on our prompts, and GPT-6.1 Sol Pro is slower and wordier again. Both went live on FastMetal on September 30, 2026, as gpt-6.1-sol and gpt-6.1-sol-pro. Existing code only needs a new model value.

The measurements

We ran 41 text tasks in Japanese and English (code review, translation, summarization, structured data extraction and more) through every model under the same conditions with our comparison runner, on September 30, 2026. Figures are medians per answer.

GPT-6.1 Sol took about twice as long as GPT-6 Sol and wrote about 20% more. The two are billed at the same per-token rates, so each request costs more by the extra output. Where speed matters most, GPT-6 Sol can still be the better choice.

Which answer is better depends on the task. Translating a Japanese product review into English, for example, the two 6.1 models opened the same sentence like this:

• GPT-6.1 Sol: "If I had one small criticism, it would be that I wish it came in a few more colors..."
• GPT-6.1 Sol Pro: "My only minor gripe is that I wish it came in a few more colors..."

All 41 answers are on the GPT-6.1 Sol and GPT-6.1 Sol Pro model pages. For a side-by-side with Claude, see GPT-6.1 Sol vs Claude Sonnet 5.5 and the comparison page.

What both models support

• A 1,050,000-token context and up to 128,000 output tokens.
• Image and file input.
• Tool calling and structured output.
• A settable reasoning effort.
• No training on your prompts. We confirmed that both models answer with training opted out.

What Sol Pro adds to every request

GPT-6.1 Sol Pro is GPT-6.1 Sol run in OpenAI's "pro" reasoning mode. The upstream adds about 1,400 input tokens to every request. In our calls through FastMetal, even a question with a one-word answer cost well over what the standard model charged for it. Most of the added tokens are billed at the cache-read rate, but Pro is still a poor fit for high volumes of short requests. In the run above, Pro also wrote about 2.3 times the output of the standard model.

A sensible order to try them in:

• Everyday chat, classification and summaries: gpt-6.1-sol
• Work where speed comes first: gpt-6-sol
• Hard problems where quality is worth the time and cost: gpt-6.1-sol-pro

How it is billed

Both models are billed per million input and output tokens, with separate rates for cache reads and cache writes. Prompts over 272,000 tokens are billed at a higher tier. GPT-6.1 Sol has the same rates as GPT-6 Sol. Since launch, Sol Pro's rates have moved slightly above the standard model's. Current rates are on the GPT-6.1 Sol and GPT-6.1 Sol Pro pages.

Billing is prepaid: you load a balance by card, from $5, and you can never spend more than you loaded. That makes a model like Pro, whose cost per request is hard to predict, safe to try with a fixed ceiling. See pricing.

Calling it

It is OpenAI-compatible, so change the model value:

curl https://api.fastmetal.ai/v1/chat/completions \
  -H "Authorization: Bearer $FASTMETAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6.1-sol",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

With the OpenAI SDK, change base_url:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.fastmetal.ai/v1",
    api_key="sk-...",
)

res = client.chat.completions.create(
    model="gpt-6.1-sol-pro",
    messages=[{"role": "user", "content": "List the failure modes in this design."}],
)
print(res.choices[0].message.content)

Everything else we serve is in the model catalog.