APIMODELSAPIMODELS

GPT-6 vs GPT-6.1: what changed in OpenAI Sol 6.1

TL;DR GPT-6.1 Sol (OpenAI Sol 6.1), released on 29 September 2026, is an upgrade to GPT-6 Sol at the same $2 / $10 list price: cached input is halved to $0.10, OpenAI reports 75.2% on DeepSWE v1.1 against 68.8% for GPT-6 Sol at about 76% lower cost per task, and the knowledge cutoff moves to 30 April 2026. The one breaking change is that reasoning_effort no longer accepts none. On APIMODELS both cost $1 / $5 per 1M tokens, so switching is a one-line model change.

When people search "GPT-6 vs 6.1" they almost always mean the Sol tier. OpenAI did not release a GPT-6.1 Astra or a GPT-6.1 Luna: on 29 September 2026 it shipped GPT-6.1 Sol, a new version of the everyday GPT-6 model, and left Astra and Luna where they were. GPT-6 Sol has not been announced for retirement, so for now both are available side by side.

This page puts the two Sol versions next to each other using OpenAI's own model pages and launch notes, then shows what the switch costs on apimodels.app. Short version: same list price, better benchmark scores, more affordable cache, one parameter to check.

Side by side

Everything below comes from OpenAI's model pages for gpt-6-sol and gpt-6.1-sol and from the 29 September launch announcement. Context window, output limit and modalities are unchanged; the differences are the cache price, the knowledge cutoff, reasoning effort and the benchmark results.

GPT-6 SolGPT-6.1 Sol
ReleasedSeptember 202629 September 2026 (DevDay)
OpenAI list (input / output)$2 / $10$2 / $10
OpenAI cached input$0.20$0.10
APIMODELS (input / output / cached)$1 / $5 / $0.10$1 / $5 / $0.05
Context / max output1.05M / 128K1.05M / 128K
Knowledge cutoff20 April 202630 April 2026
reasoning_effortnone to max (6 levels)low to max (5 levels, no none)
DeepSWE v1.1 (OpenAI)68.8% best75.2% at high
AutomationBench (OpenAI)4.8 points lower31.7% at medium

What the benchmarks say

OpenAI's headline number is coding: GPT-6.1 Sol scores 75.2% on DeepSWE v1.1 at high reasoning effort, above GPT-6 Sol's best of 68.8%, and OpenAI says it gets there at roughly 76% lower cost per task. On AutomationBench it reaches 31.7% at medium effort, 4.8 points above GPT-6 Sol. On OSWorld 2.0 it scores 71.4%, close to GPT-6 Astra's 73.5% at roughly one seventh of Astra's cost per task.

These are OpenAI's figures, not ours. The practical reading: for agentic coding and multi-step automation, 6.1 should finish more tasks with fewer tokens, which is where the lower cost per task comes from. For short chat or extraction work the difference will be smaller.

Switching from GPT-6 to GPT-6.1

Change the model field from gpt-6-sol to gpt-6.1-sol. Nothing else in the request has to change, with one exception: if your code sends reasoning_effort: "none", switch it to "low", because 6.1 does not accept none. The bare alias gpt-6.1 resolves to gpt-6.1-sol, while gpt-6 still resolves to GPT-6 Sol, so existing integrations are never upgraded behind your back.

Cost does not go up. Both are $1 / $5 per 1M tokens on APIMODELS, half of OpenAI's $2 / $10 list, and 6.1's cached input is $0.05 instead of $0.10, so long system prompts and repeated context get more affordable. Above 272K input tokens both bill input at 2× and output at 1.5× for the whole request.

Quick start

cURL

curl https://api.apimodels.app/v1/chat/completions \
  -H "Authorization: Bearer $APIMODELS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6.1-sol",
    "reasoning_effort": "low",
    "messages": [{"role": "user", "content": "Summarize the changes between two versions of this API in three bullets."}]
  }'

Python

from openai import OpenAI
import os

client = OpenAI(api_key=os.environ["APIMODELS_API_KEY"], base_url="https://api.apimodels.app/v1")

resp = client.chat.completions.create(
    model="gpt-6.1-sol",          # was "gpt-6-sol"
    reasoning_effort="low",       # 6.1 has no "none"; use "low" for the most affordable run
    messages=[{"role": "user", "content": "Write a unit test for this function."}],
)
print(resp.choices[0].message.content)

FAQ

What is the difference between GPT-6 and GPT-6.1?

GPT-6.1 is an update to the Sol tier only. GPT-6.1 Sol keeps GPT-6 Sol's $2 / $10 list price, 1.05M context and 128K output, halves cached input to $0.10, moves the knowledge cutoff to 30 April 2026, drops the none level from reasoning_effort, and scores higher on OpenAI's benchmarks — 75.2% vs 68.8% on DeepSWE v1.1.

GPT 6 vs 6.1: which one should I use?

Use GPT-6.1 Sol for new work. It costs the same as GPT-6 Sol on apimodels.app ($1 / $5 per 1M tokens), has more affordable cached input ($0.05 vs $0.10) and scores higher on OpenAI's coding and automation benchmarks. Stay on GPT-6 Sol only if your code depends on reasoning_effort: none and you cannot change it yet.

What is OpenAI Sol 6.1?

"Sol 6.1" is how many people search for GPT-6.1 Sol, the model OpenAI released on 29 September 2026. Its API model id is gpt-6.1-sol. On apimodels.app it costs $1 per 1M input tokens and $5 per 1M output, half the official price.

Is GPT-6.1 Sol better than GPT-6 Sol?

On OpenAI's published benchmarks, yes: 75.2% vs 68.8% on DeepSWE v1.1 and 4.8 points higher on AutomationBench (31.7%), at about 76% lower cost per coding task. It costs the same per token, so for new work there is little reason to pick GPT-6 Sol.

How much does GPT-6.1 Sol cost through the API?

OpenAI lists it at $2 per 1M input tokens, $10 per 1M output and $0.10 cached. On apimodels.app it is $1 / $5 / $0.05 — half the list — with the same 2× input and 1.5× output rule above 272K input tokens. No OpenAI account is needed.

Will my gpt-6 requests be upgraded to 6.1 automatically?

No. On apimodels.app the alias gpt-6 keeps pointing to GPT-6 Sol. Use gpt-6.1-sol, or the alias gpt-6.1, when you want the new version.

Does GPT-6.1 Sol support reasoning_effort none?

No. OpenAI's model page lists low, medium (default), high, xhigh and max. If you migrate code that sends none, change it to low.

Related

등재된 곳
APIMODELS on Product HuntAPIMODELS featured on ToolifyAPIMODELS featured on There's An AI For That