GPT-5.6 API Pricing: Sol vs Terra vs Luna Rates, Promo End, and Which to Use
GPT-5.6 Sol, Terra and Luna API prices in every mode, the November 30 end of Sol's promotional rate, cache writes, the 272K long-context step, worked cost examples, a Sol vs Terra vs Luna routing guide, and when to move to GPT-6.1.
OpenAI released the GPT-5.6 family on July 9, 2026 with three API tiers: Sol for the hardest work, Terra as the balanced production model and Luna for high-volume routing. GPT-6 has since replaced it as OpenAI’s current generation, but GPT-5.6 is still on the price list and still runs in many production systems. This guide covers its full pricing and how to choose between the three tiers. For all OpenAI models on one page, see the OpenAI API pricing overview.
Checked against OpenAI’s pricing page on November 16, 2026:
| Model | API model ID | Input | Cached input | Output |
|---|---|---|---|---|
| GPT-5.6 Sol | gpt-5.6-sol | $4.00 | $0.40 | $20.00 |
| GPT-5.6 Terra | gpt-5.6-terra | $2.00 | $0.20 | $12.00 |
| GPT-5.6 Luna | gpt-5.6-luna | $0.20 | $0.02 | $1.20 |
USD per 1M tokens, Standard processing, prompts up to 272K tokens. The unsuffixed gpt-5.6 alias routes to Sol.
Sol’s $4 / $20 rate is promotional and ends on November 30, 2026. OpenAI has not published the rate that applies afterwards. Before the promotion, Sol was listed at $5 / $30; if you have to budget December now, use that figure as the cautious case and recheck the pricing page on December 1.
Models at a Glance
All three models have a 1,050,000-token context window, 128,000 maximum output tokens and a February 16, 2026 knowledge cutoff. They accept text and images and return text. Streaming, function calling, structured outputs and reasoning are supported; fine-tuning is not.
| Model | Positioning | Reasoning effort |
|---|---|---|
| GPT-5.6 Sol | Hard professional, coding, research and agent work | none through max |
| GPT-5.6 Terra | Everyday production work at lower cost | none through max |
| GPT-5.6 Luna | High-volume routing and lighter workloads | none through max |
Pro mode is not a separate gpt-5.6-pro model ID. Keep the selected GPT-5.6 model and set reasoning.mode to pro.
Full Price Table
A cached read is 10% of base input. Writing to the cache is billed separately: 1.25x base input for 30-minute retention, 2.5x for 12-hour retention. For prompts over 272K input tokens, the entire request moves to long-context rates: 2x input, cached input and cache writes, 1.5x output.
| Model | Mode | Input | Cached | Cache write (30 min) | Output |
|---|---|---|---|---|---|
| Sol | Standard ≤272K | $4.00 | $0.40 | $5.00 | $20.00 |
| Sol | Standard >272K | $8.00 | $0.80 | $10.00 | $30.00 |
| Sol | Batch / Flex ≤272K | $2.00 | $0.20 | $2.50 | $10.00 |
| Sol | Fast ≤272K | $8.00 | $0.80 | $10.00 | $40.00 |
| Terra | Standard ≤272K | $2.00 | $0.20 | $2.50 | $12.00 |
| Terra | Standard >272K | $4.00 | $0.40 | $5.00 | $18.00 |
| Terra | Batch / Flex ≤272K | $1.00 | $0.10 | $1.25 | $6.00 |
| Terra | Fast ≤272K | $4.00 | $0.40 | $5.00 | $24.00 |
| Luna | Standard ≤272K | $0.20 | $0.02 | $0.25 | $1.20 |
| Luna | Standard >272K | $0.40 | $0.04 | $0.50 | $1.80 |
| Luna | Batch / Flex ≤272K | $0.10 | $0.01 | $0.125 | $0.60 |
| Luna | Fast ≤272K | $0.40 | $0.04 | $0.50 | $2.40 |
Long-context Batch and Flex rates are half the long-context Standard rates (Sol $4 / $0.40 / $15, Terra $2 / $0.20 / $9, Luna $0.20 / $0.02 / $0.90). Fast mode, called Priority until July 30, 2026, is 2x Standard. Regional processing endpoints add 10%, and web search is billed per call on top of tokens.
Worked Cost Examples
Example 1: The same monthly workload on each tier
10M uncached input tokens and 2M output tokens a month, Standard, short context:
| Model | Input | Output | Total |
|---|---|---|---|
| Sol | 10 × $4 = $40.00 | 2 × $20 = $40.00 | $80.00 |
| Terra | 10 × $2 = $20.00 | 2 × $12 = $24.00 | $44.00 |
| Luna | 10 × $0.20 = $2.00 | 2 × $1.20 = $2.40 | $4.40 |
Terra costs 55% of Sol here; Luna costs 5.5%. Real task costs can differ, because the tiers may use different numbers of reasoning and output tokens to finish the same job.
Example 2: A reused 1M-token prefix
Write a 1M-token prefix to the cache once (30-minute retention) and read it nine more times, compared with sending it uncached ten times:
| Model | Cached | Uncached |
|---|---|---|
| Sol | $5 + 9 × $0.40 = $8.60 | 10 × $4 = $40.00 |
| Terra | $2.50 + 9 × $0.20 = $4.30 | 10 × $2 = $20.00 |
| Luna | $0.25 + 9 × $0.02 = $0.43 | 10 × $0.20 = $2.00 |
Note that a 1M-token prefix is above 272K, so in a real request these input lines would bill at long-context rates (double the figures above). The ratio between cached and uncached stays the same: caching saves 78.5%.
Example 3: After the Sol promotion ends
The Example 1 workload on Sol costs $80 at the promotional rate. At the pre-promotion $5 / $30 it would cost 10 × $5 + 2 × $30 = $110. The same workload on GPT-6.1 Sol ($2 / $10) costs 10 × $2 + 2 × $10 = $40, and on GPT-6.1 Terra ($1 / $5) 10 × $1 + 2 × $5 = $20.
Sol vs Terra vs Luna: Which to Use
Choose GPT-5.6 Sol when:
- a failed attempt costs more than the price difference between tiers;
- the task involves difficult debugging, architecture, security, science or long-running agent work;
- you need
maxeffort orproreasoning mode.
Choose GPT-5.6 Terra when:
- you need a strong default for coding agents, document work and production assistants;
- Sol is too expensive for every request but a small model is too risky;
- you want the same context and output limits as Sol at half the input price and 40% less output price.
Choose GPT-5.6 Luna when:
- the work is classification, extraction, summarization, search triage or simple code changes;
- requests are high-volume and can be escalated when confidence is low;
- you need the full 1.05M window without Sol-level reasoning on every call. Luna has the same context and output limits as Sol and Terra.
A practical routing policy
- Start routine requests on Luna.
- Route normal coding and professional work to Terra.
- Escalate ambiguous, high-impact or failed tasks to Sol.
- Send asynchronous traffic through Batch or Flex.
- Keep reusable prefixes stable so reads hit the cache.
Should You Still Use GPT-5.6?
On price alone, no new project should start on GPT-5.6. Each GPT-6 replacement is cheaper per token:
| GPT-5.6 | GPT-6 replacement | Price change |
|---|---|---|
| Sol, $4 / $20 (promo) | GPT-6.1 Sol, $2 / $10 | Half |
| Terra, $2 / $12 | GPT-6.1 Terra, $1 / $5 | Half the input, 58% less output |
| Luna, $0.20 / $1.20 | GPT-6 Luna, $0.10 / $0.50 | Half the input, 58% less output |
Keep GPT-5.6 only where prompts and evals are tuned to it and a migration has not been tested yet. OpenAI’s migration advice for earlier generations was to test the same reasoning effort and one level lower; apply the same approach when moving to GPT-6.1. Detailed GPT-6 rates are in the GPT-6 pricing guide.
How GPT-5.6 Prices Changed
| Date | Change |
|---|---|
| Jul 9, 2026 | Launch: Sol $5 / $30, Terra $2.50 / $15, Luna $1 / $6 |
| Jul 30, 2026 | Priority processing renamed Fast mode |
| Aug 1, 2026 | Terra cut to $2 / $12; Luna cut 80% to $0.20 / $1.20 |
| Sep 2026 | Sol promotional price $4 / $20 |
| Nov 30, 2026 | Sol promotion ends |
Sources and dates are logged in the AI API price change tracker. For GPT-5.6 inside Codex, which bills in credits rather than tokens, see Codex pricing and limits.
API Example
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-5.6-terra",
reasoning={"effort": "medium"},
input="Compare these two API designs and return the three highest-risk tradeoffs."
)
print(response.output_text)
Use the AI Model Pricing Calculator to test your own input, output and cache mix.
Sources
Checked November 16, 2026:
- OpenAI API pricing: all GPT-5.6 rates, cache-write retention prices, long-context and Batch/Flex/Fast rates, promotion end date
- GPT-5.6 Sol model page
- GPT-5.6 Terra model page
- GPT-5.6 Luna model page
- GPT-5.6 launch announcement
- DevTk.AI price change tracker: launch prices and later changes
Next step
Related Posts
GPT-6 API Pricing 2026: Astra, GPT-6.1 Sol, GPT-6.1 Terra and Luna in Every Mode
2026-09-25
The August 2026 AI API Price War: Google Discounts as DeepSeek Raises Prices
2026-08-01
Kimi K3 vs GPT-5.6 Sol: API Price, 1M Context, Coding, and Caching
2026-07-19