What Changed Between GPT-5.4 and GPT-5.6?
GPT-5.6 is the same shape of model line as GPT-5.4 at 20% less on the mid tier and 3.75 times less on the small tier. The flagship moved somewhere else entirely, to gpt-6-astra at $10.00 input and $50.00 output per million tokens.
Three things changed, and only one of them is the number on the invoice.
The names changed first. In July 2026 OpenAI retired the Instant, Thinking, mini and nano labels for Sol, Terra and Luna. So gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna are tiers of one release, not three products.
The mid tier got cheaper. gpt-5.4 at $2.50/$15.00 gives way to gpt-5.6-terra at $2.00/$12.00.
The small tier collapsed. gpt-5.4-mini at $0.75/$4.50 gives way to gpt-5.6-luna at $0.20/$1.20.
Every figure below was checked against the vendors' own documentation on 16 September 2026. AI Perks tracks $7.7M in credits across 194 companies that offset exactly these bills.
GPT-5.4 vs GPT-5.6: Every Current Price
All figures are USD per million tokens, sorted by combined cost. The right-hand column is our arithmetic on the listed prices, not a vendor figure.
| OpenAI model | Input / 1M | Output / 1M | 1M in + 1M out (est.) |
|---|---|---|---|
| gpt-6-astra | $10.00 | $50.00 | $60.00 |
| gpt-5.5 | $5.00 | $30.00 | $35.00 |
| gpt-5.6-sol | $4.00 | $20.00 | $24.00 |
| gpt-5.4 | $2.50 | $15.00 | $17.50 |
| gpt-5.6-terra | $2.00 | $12.00 | $14.00 |
| gpt-5.4-mini | $0.75 | $4.50 | $5.25 |
| gpt-5.4-nano | $0.20 | $1.25 | $1.45 |
| gpt-5.6-luna | $0.20 | $1.20 | $1.40 |
Output costs 6x input on most OpenAI tiers. gpt-5.4, terra, luna, mini and gpt-5.5 all run a 6x multiple, with gpt-5.4-nano at 6.25x and gpt-5.6-sol at 5x. Your bill is driven by how verbose your completions are, not by how much context you stuff in.
gpt-5.5 costs more than gpt-5.6-sol. At $5.00/$30.00 it runs 46% above sol per million in and out.
One threshold to plan around: gpt-6-astra charges $20.00/$75.00 above 200,000 tokens of context. On a 50M in / 10M out month that is $1,750 instead of $1,000, a 75% premium.
What Happened to GPT-5.4 mini?
gpt-5.4-mini at $0.75/$4.50 is undercut by exactly 3.75 times on both input and output by gpt-5.6-luna at $0.20/$1.20. There is no listed price argument left for running mini on new work.
The ratio is identical on both sides, which makes the comparison unusually clean. A workload costing $5.25 per million in and out on mini costs $1.40 on luna, a 73% cut.
Luna also swallows the old floor. gpt-5.4-nano lists at $0.20/$1.25, so luna matches it on input and is marginally cheaper on output. One tier now covers the bottom of OpenAI's range where two sat before.
gpt-5.4-mini and gpt-5.4-nano are both still on OpenAI's current price list and still billable. The point is not that they are going away, it is that they cost 3.75x and 1.04x more than the tier already in front of you.
If the small tier is where your volume lives, credits matter more than the rate card. AI Perks covers model, cloud and infrastructure credits.
What These Models Cost at Real Volume
At 50 million input and 10 million output tokens a month, gpt-5.4 costs about $275 and gpt-5.6-terra about $220. These are estimates derived from the listed prices, with no discounts, batching or caching applied.
| Model | Input cost | Output cost | Monthly total (est.) |
|---|---|---|---|
| gpt-6-astra | $500.00 | $500.00 | $1,000.00 |
| gpt-5.5 | $250.00 | $300.00 | $550.00 |
| gpt-5.6-sol | $200.00 | $200.00 | $400.00 |
| gpt-5.4 | $125.00 | $150.00 | $275.00 |
| gpt-5.6-terra | $100.00 | $120.00 | $220.00 |
| gpt-5.4-mini | $37.50 | $45.00 | $82.50 |
| gpt-5.6-luna | $10.00 | $12.00 | $22.00 |
The gap between the top and bottom of one vendor's current line is 45x at this volume. Tier choice is a bigger lever than vendor choice, and far bigger than prompt trimming.
A year of mid-tier inference here is a few thousand dollars, well inside what one credit grant covers. AI Perks is where to find them.
How OpenAI Compares to Anthropic in September 2026
Claude Sonnet 5 at $2/$10 currently undercuts gpt-5.6-terra at $2/$12. Identical input price, 17% cheaper output, and $12.00 against $14.00 per million in and out on our arithmetic.
| Anthropic model | Input / 1M | Output / 1M | 1M in + 1M out (est.) |
|---|---|---|---|
| Claude Fable 5.1 | $10.00 | $50.00 | $60.00 |
| Claude Mythos 5.1 | $10.00 | $50.00 | $60.00 |
| Claude Opus 5 | $5.00 | $25.00 | $30.00 |
| Claude Sonnet 5 | $2.00 | $10.00 | $12.00 |
| Claude Haiku 4.5 | $1.00 | $5.00 | $6.00 |
Here is the part almost nobody has written up. Anthropic had scheduled Sonnet 5 to rise from $2/$10 to $3/$15 on 1 September 2026, and cancelled it. Its pricing page states that the increase "will not occur." The price announced as introductory through 31 August is now simply the standard price.
That cancellation flipped who wins the mid tier. At the scheduled $3/$15, Sonnet 5 would have cost $18.00 per million in and out against terra's $14.00, making OpenAI 22% cheaper. At the price that stands, Sonnet 5 is 14% cheaper than terra instead. A price rise that never happened changed the answer to "which mid tier is cheapest" for the whole quarter.
One structural note. Anthropic runs a 5x output multiple across its entire line against OpenAI's 6x on most tiers, so Anthropic gets relatively cheaper the more your workload generates rather than reads. Claude Opus 5, launched 24 July 2026, holds the same $5/$25 as Opus 4.8.
Meta prices Muse Spark 1.1 through 1.3 at $1.25/$4.25, a 3.4x output multiple and the narrowest here. Its Contributor tier drops to $0.10/$0.20, with the explicit trade that Meta trains on your prompts and completions. Meta runs no startup credit program for Muse, so that bill is yours in full.
Two deliberate omissions. xAI token prices are not printed here because published sources conflict badly, so price Grok in xAI's own console. Gemini is left out because its prices were not verified for this page, and a guessed number is worth less than none.
The December Shutdown Is a Price Rise for Some Teams
OpenAI shuts down the gpt-5, gpt-5-mini and gpt-5-nano API on 10 December 2026. Do not start new builds on them, and budget for the migration, because for two of the three it is an increase.
| Legacy model | Price | Nearest current tier | Change per 1M in + out (est.) |
|---|---|---|---|
| gpt-5 ($1.25/$10.00) | $11.25 | gpt-5.6-terra ($14.00) | +24% |
| gpt-5-mini ($0.25/$2.00) | $2.25 | gpt-5.6-luna ($1.40) | -38% |
| gpt-5-nano ($0.05/$0.40) | $0.45 | gpt-5.6-luna ($1.40) | +211% |
The legacy line ran an 8x output multiple against terra and luna's 6x, which is why the change differs so much by tier. Teams on gpt-5-nano face roughly triple the token cost, and that is the migration nobody has costed yet.
Which Tier Fits Your Workload
Pick on output volume and latency tolerance first, then check the input price. Output runs 5 to 6 times the input rate at OpenAI and Anthropic, so it decides the bill.
| If your workload is | Start with | Est. cost per 1M in + out |
|---|---|---|
| High-volume classification, routing, extraction | gpt-5.6-luna | $1.40 |
| Product features, chat, summarisation | gpt-5.6-terra or Claude Sonnet 5 | $14.00 / $12.00 |
| Heavy reasoning, agents, long tool chains | gpt-5.6-sol or Claude Opus 5 | $24.00 / $30.00 |
| Frontier quality, cost secondary | gpt-6-astra or Claude Fable 5.1 | $60.00 / $60.00 |
| Anything still on gpt-5.4-mini | gpt-5.6-luna | 3.75x cheaper |
Capability is the one variable this page does not cover. No benchmark scores, context windows or quality rankings appear here, because they were not verified. An unverified capability claim is what turns a cheap tier into an expensive one.
Test two tiers on your own evaluation set first. Luna and terra are 10x apart, which buys real tolerance for a slightly worse answer.
Then get the bill covered. AI Perks tracks $7.7M in credits across 194 companies, including every model provider priced above.
Frequently Asked Questions
Is GPT-5.6 cheaper than GPT-5.4?
Yes. gpt-5.6-terra lists at $2.00 input and $12.00 output per million tokens against gpt-5.4 at $2.50 and $15.00, a flat 20% cut on both sides. Processing one million input and one million output tokens costs about $14.00 on terra versus $17.50 on gpt-5.4, based on OpenAI's own listed prices.
How much does GPT-5.4 mini cost?
gpt-5.4-mini lists at $0.75 per million input tokens and $4.50 per million output tokens. Its replacement gpt-5.6-luna lists at $0.20 and $1.20, exactly 3.75 times cheaper on both sides. Workloads still running on mini pay a premium for no listed advantage. Credits that offset it are tracked at AI Perks.
What are Sol, Terra and Luna?
They are OpenAI's tier names, introduced in July 2026 to replace the older Instant, Thinking, mini and nano labels. Sol is the $4.00/$20.00 tier, Terra is $2.00/$12.00, and Luna is $0.20/$1.20, the cheapest current OpenAI tier. The flagship above all three is gpt-6-astra at $10.00/$50.00.
Did Anthropic raise Claude Sonnet 5 pricing?
No. Anthropic had scheduled Sonnet 5 to move from $2/$10 to $3/$15 per million tokens on 1 September 2026, then cancelled it. Its pricing page states the increase "will not occur," so $2/$10 is now the standard price rather than an introductory one. That keeps Sonnet 5 cheaper than gpt-5.6-terra on output.
When do the older GPT-5 models shut down?
OpenAI has set 10 December 2026 as the API shutdown date for gpt-5, gpt-5-mini and gpt-5-nano, so do not start new builds on them. Teams on gpt-5-nano at $0.05/$0.40 face the sharpest jump, since the nearest current tier costs roughly three times as much per million tokens.
Which model is cheapest for high output volume?
Among the prices verified here, gpt-5.6-luna at $1.20 per million output tokens is the cheapest current OpenAI tier, and Claude Haiku 4.5 at $5.00 is Anthropic's. Meta Muse Spark runs $4.25 output on $1.25 input, the narrowest multiple of any provider listed. Credits covering all of them sit at AI Perks.
Pick the right tier, then stop paying for it.