Short answer
GPT-6 Sol is available on Moyi API as gpt-6-sol with a 272,000-token context window. OpenAI's list price is $2 per million input tokens and $10 per million output; on Moyi API, as of the publish date, it is about $0.30 / $1.50. To run Codex CLI on Sol you need one ~/.codex/config.toml with four lines and your key in OPENAI_API_KEY. You pay per token, by card or USDT, with no subscription.
Sol, Astra, or Luna: which one to pick
On September 23, 2026 OpenAI opened the GPT-6 line with three models. All three are on Moyi API under one key; they differ in price and in what they are built for.
| Model | ID | List price (input / output, per 1M) | Built for |
|---|---|---|---|
| GPT-6 Astra | gpt-6-astra |
$10 / $50 | Flagship: hardest reasoning, large agentic tasks |
| GPT-6 Sol | gpt-6-sol |
$2 / $10 | Workhorse: everyday coding, refactors, review, agents in CI |
| GPT-6 Luna | gpt-6-luna |
$0.10 / $0.50 | Fast and cheap: completions, classification, small edits |
Sol is the sensible default for Codex CLI and most agents: five times cheaper than Astra with the same 272k context (roughly 8–10k lines of code plus docs in one request). Switch to Astra for a specific hard task; push high-volume cheap work to Luna.
Pricing on Moyi API
Per 1 million tokens, in USD:
| OpenAI list price | Moyi API | |
|---|---|---|
| Input | $2.00 | ≈ $0.30 |
| Output | $10.00 | ≈ $1.50 |
| Cache read | $0.20 | ≈ $0.03 |
| Cache write | $2.50 | ≈ $0.37 |
The Moyi API figure tracks the upstream price list, so check the pricing page for the live number; the table shows the publish-date values.
In session terms: a typical Codex task on a mid-size project reads about 150k input tokens and writes about 15k. At list price that is $0.30 + $0.15 = $0.45; on Moyi API roughly $0.05 + $0.02 = $0.07. Repeat requests against the same code are billed at the cache-read rate, cheaper still — see Billing FAQ for how caching is charged.
Set up Sol in Codex CLI
You need a Moyi API API key — create one in the console. Codex CLI installs with npm install -g @openai/codex; the full walkthrough for macOS, Windows, and Linux is in the Codex guide.
Configuration
macOS and Linux — run in a terminal:
mkdir -p ~/.codex && cat > ~/.codex/config.toml <<'EOF'
model_provider = "teamorouter"
model = "gpt-6-sol"
model_reasoning_effort = "high"
[model_providers.teamorouter]
name = "Moyi API"
base_url = "https://api.moyiapi.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"
EOF
echo 'export OPENAI_API_KEY="your Moyi API key"' >> ~/.zshrc && source ~/.zshrc
Windows — in PowerShell:
New-Item -ItemType Directory -Force "$HOME\.codex" | Out-Null
@'
model_provider = "teamorouter"
model = "gpt-6-sol"
model_reasoning_effort = "high"
[model_providers.teamorouter]
name = "Moyi API"
base_url = "https://api.moyiapi.com/v1"
env_key = "OPENAI_API_KEY"
wire_api = "responses"
'@ | Set-Content -Path "$HOME\.codex\config.toml" -Encoding UTF8
setx OPENAI_API_KEY "your Moyi API key"
Open a new terminal after setx. Run codex — the first line of the reply identifies the model as gpt-6-sol.
Switching inside a session
/model in Codex lists the available models: jump to gpt-6-astra for a hard task and back to gpt-6-sol. model_reasoning_effort (low / medium / high) controls how much the model thinks before answering — medium is enough for routine edits and is noticeably faster and cheaper.
Other agents
The same key and base_url work in any agent with an OpenAI-compatible provider: OpenCode, Cline for VS Code, DeepSeek Harness, Cursor. Only the model name changes — set it to gpt-6-sol.
Calling the API directly
Chat Completions — the format every OpenAI SDK understands:
from openai import OpenAI
client = OpenAI(base_url="https://api.moyiapi.com/v1", api_key="YOUR_KEY")
resp = client.chat.completions.create(
model="gpt-6-sol",
messages=[{"role": "user", "content": "Find the bug in this function: ..."}],
)
print(resp.choices[0].message.content)
Responses API — the format Codex itself uses; available for GPT models:
curl https://api.moyiapi.com/v1/responses \
-H "Authorization: Bearer YOUR_KEY" \
-H "content-type: application/json" \
-d '{
"model": "gpt-6-sol",
"input": "Write a migration that adds an email column to the users table"
}'
Full parameter reference: API integration.
Paying for usage
You top up a balance in the console and are charged only for tokens actually used — no subscription, no monthly fee.
- Card — Visa, Mastercard and other major cards via Stripe.
- USDT — choose "Pay with crypto" in the top-up dialog, send USDT to the address shown; the balance is credited within one to five minutes.
No minimum bundle, nothing expires: top up $10 and you spend exactly $10 at the prices in the table.
FAQ
Codex says model gpt-6-sol was not found. Check that config.toml has model_provider = "teamorouter" and that the key is set in OPENAI_API_KEY in the shell that launched Codex. GET /v1/models with your key returns the available list.
How is gpt-6-sol different from gpt-5.6-sol? Different generations: 5.6 Sol is the previous line, 6 Sol the current one. If your config.toml still says gpt-5.6-sol, change it by hand — Codex will not upgrade the model on its own.
Does Fast mode work? For GPT-5.6 Sol it was enabled with service_tier = "fast" in config.toml. For GPT-6 Sol, check the pricing page: if the mode ships for this line, it will have its own price there.
What are cache read and cache write? When the same long context (your repository) is sent again, it is served from cache at the read rate — several times cheaper than normal input. The first send is billed as a write. Details in the billing FAQ.
Are there request limits? Yes, per model — see the rate limits page.
Next steps
- Install and configure Codex CLI — if you're starting from scratch.
- Claude Opus 5.5 via API — the other model released the same day.
- Create an API key — and run Sol in Codex in five minutes.