GPT-6.1 Sol vs GPT-6 Sol: same price, cache reads at half
Short answer: GPT-6.1 Sol and GPT-6 Sol have the same input and output prices ($2 / $10 per million tokens). The only list-price change is cached input, down from $0.20 to $0.10. For agents with long, well-cached context, moving to 6.1 costs the same or less; when you migrate, watch reasoning effort and a few sampling parameters.
gpt-6.1-solOpenAI · 1.05M · Stable $2.00 / $10.00Model page →USD per million tokens
A request whose input (cached tokens included) passes the threshold is billed entirely at the long-context prices.
A dash means no price is recorded, not free usage. The request receipt is the final billing record.
gpt-6-sol (previous)
gpt-6-solOpenAI · 1.05M · Stable $2.00 / $10.00 · Discount $0.36 / $1.80Model page →USD per million tokens
A request whose input (cached tokens included) passes the threshold is billed entirely at the long-context prices.
A dash means no price is recorded, not free usage. The request receipt is the final billing record.
Prices side by side #
| Item | GPT-6.1 Sol | GPT-6 Sol |
|---|---|---|
| Input | $2.00 | $2.00 |
| Cached input | $0.10 | $0.20 |
| Cache write | $2.50 | $2.50 |
| Output | $10.00 | $10.00 |
| Input over 272K: input / cached / output | $4.00 / $0.20 / $15.00 | $4.00 / $0.40 / $15.00 |
USD per million tokens. Both have a 1,050,000-token context window (up to 922,000 input per request) and 128,000 maximum output.
Mind the 272K line: a request with more than 272K input tokens is billed at the long-context price for the whole request, not only the tokens above the line.
What it means for a long agent session #
Say a coding agent runs 50 turns. Each turn carries an 80K-token fixed prefix (system prompt, tool definitions, files already read, all cached), 5K new tokens, and 2K of output:
| GPT-6 Sol | GPT-6.1 Sol | |
|---|---|---|
| Per turn | 80K × 0.20 + 5K × 2 + 2K × 10 ≈ $0.046 | 80K × 0.10 + 5K × 2 + 2K × 10 ≈ $0.038 |
| Whole session (50 turns) | ≈ $2.30 | ≈ $1.90 |
About 17% less. The longer the cached prefix, the bigger the saving.
Now long context: one 300K-token request (280K cached, 20K new) with 3K output:
| GPT-6 Sol | GPT-6.1 Sol | |
|---|---|---|
| 300K (over 272K) | 280K × 0.40 + 20K × 4 + 3K × 15 ≈ $0.237 | 280K × 0.20 + 20K × 4 + 3K × 15 ≈ $0.181 |
| Kept under 250K | — | 230K × 0.10 + 20K × 2 + 3K × 10 ≈ $0.093 |
Keeping context under 272K saves more than switching models does.
What to change when you migrate #
From OpenAI's migration guide:
- Change the model ID from
gpt-6-soltogpt-6.1-sol. - Reasoning effort: 6.1 supports low, medium (default), high, xhigh and max.
noneandminimalare gone. Replacenonewithlow; if you usedminimal, start atlowand compare. - Remove sampling parameters:
temperature,top_p,top_logprobsandlogprobsare no longer supported. - Use the Responses API for tool calling. Chat Completions works only for requests without tools.
- To change effort mid-conversation, send a
configuration_updateitem and leave the request-level effort alone, so caching keeps working. - Coming from GPT-5.5 or earlier: replace
prompt_cache_retentionwithprompt_cache_options.ttlset to"30m".
GPT-6 Sol has no retirement date, so there is no rush, but new projects should start on 6.1.
On Lazu #
Both models work with the same Lazu key, on Chat Completions and Responses, and in Codex (see Codex with third-party models).
GPT-6 Sol also has a discount lane on Lazu that accepts only Codex client requests. GPT-6.1 Sol is on the stable lane only, at OpenAI's official price.
from openai import OpenAI
client = OpenAI(base_url="https://api.lazu.ai/v1", api_key="YOUR_LAZU_KEY")
resp = client.responses.create(
model="gpt-6.1-sol",
reasoning={"effort": "low"},
input="In one sentence, how does GPT-6.1 Sol's pricing differ from GPT-6 Sol's?",
)
print(resp.output_text)
The console's usage log shows each request's cached tokens and actual charge.