GPT API Price Cut: 5 Models Now Cost 20% Less on reAPI

chatgpt api pricing

ChatGPT API pricing means usage-based GPT model charges, not a fixed ChatGPT subscription price. For the six GPT-6 and GPT-5.6 tiers checked on October 23, 2026, OpenAI Standard uncached input ranges from $0.10 to $10 per million tokens, and output from $0.50 to $50 in the shorter-prompt tier. reAPI's observed rates for those same categories range from $0.08 to $8 input and $0.40 to $40 output. These are selected models, not a complete GPT catalog^[7][8].

All figures in this overview are USD per 1 million tokens, for requests with no more than 272,000 input tokens. Cached input is a separate category; the ordinary input rate is not added to it. Output includes billable reasoning tokens. The table excludes tool charges, taxes, regional modifiers, and special processing modes^[7][9].

The ordinary input/output rates above are 20% lower on reAPI, but that is not a universal discount across every billing category. GPT-6 Sol's displayed short-context cache read is $0.016, not the $0.160 implied by applying 80% to OpenAI's $0.20. The published long-context cache-read rate for GPT-6 Luna is $0.160 on reAPI versus $0.02 at OpenAI. Compare the actual category, not a blanket percentage^[7][8].

For an uncached request with 20,000 input tokens and 5,000 output tokens, GPT-6 Astra costs (20,000 × $10 + 5,000 × $50) / 1,000,000 = $0.45 at OpenAI Standard, or (20,000 × $8 + 5,000 × $40) / 1,000,000 = $0.36 at the observed reAPI rates. This is price arithmetic, not a measured invoice. With caching, use disjoint token categories instead:

cost = (ordinary input × input rate
      + cache read × read rate
      + 30m cache write × 30m write rate
      + 12h cache write × 12h write rate
      + output including reasoning × output rate) / 1,000,000
ordinary input = total input - cache read - cache write

The two cache-write lifetimes above describe OpenAI's published billing. For these six models, its default 30-minute write rate is 1.25× ordinary input and its 12-hour write rate is 2.5×; each replaces ordinary input for those tokens. The reAPI pricing displays do not identify a write lifetime, so do not assume they offer the same 12-hour control. Above 272,000 input tokens, OpenAI reprices the whole request at twice the input/cache rates and 1.5× output. See the GPT-6 Astra pricing guide and GPT-5.6 tier comparison for the full conditions rather than applying short-context prices to a long prompt^[7][9].

ChatGPT subscriptions and API usage have separate billing systems. A ChatGPT plan price is not an API token allowance; paid API usage is set up in the API account separately. reAPI requests use a reAPI key and its own credit balance. Identify the account paying for the request before estimating costs^[10][11]. The five-model price-cut comparison below retains its October 22, 2026 snapshot; its GPT-5.4 and GPT-5.5 rows have not been relabeled as a new October 23 verification.

The GPT API price cut recorded below is easy to misunderstand, because two different prices moved. OpenAI lowered its own list prices for the GPT-5.6 family after this article first ran, while GPT-5.4 and GPT-5.5 kept theirs.^[5] Separately, the October 22 comparison lists all five selected GPT models below OpenAI's standard short-context input and output token rates: four of them at 20% below, and GPT-5.4 at 60% below. The five-model comparison was checked on October 22, 2026.

The discount covers GPT-5.4, GPT-5.5, and the complete GPT-5.6 family—Sol, Terra, and Luna. The SDK example below uses Chat Completions. Check the model-specific endpoint and key: GPT-5.6 uses the unified reAPI workspace, while GPT-5.4 and GPT-5.5 use the separate legacy gateway. This does not establish compatibility with OpenAI Responses or every tool feature.^[11]

This guide compares every rate, calculates the savings on a real workload, and explains when OpenAI's Batch API or prompt caching can still change the answer.

TL;DR

• All five models in the October 22 comparison are below OpenAI's standard short-context input/output prices on reAPI: GPT-5.5, GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna at 20% below, and GPT-5.4 at 60% below.
• OpenAI's own GPT-5.6 list prices also fell. Sol is now $4/$20, Terra $2/$12, and Luna $0.20/$1.20 per million tokens; OpenAI labels Sol's rate promotional through at least November 21, 2026.^[5]
• The cheapest of those five models is GPT-5.6 Luna at $0.16 input and $0.96 output per million tokens on reAPI.
• GPT-5.4 is now cheaper than Terra on reAPI: $1 input and $6 output per million tokens, against $1.60 and $9.60. Benchmark both before choosing.
• GPT-5.6 Sol is now cheaper than GPT-5.5: $3.20 input and $16 output per million tokens on reAPI, against $4 and $24.
• A 20% token discount produces a 20% bill reduction when request mix and token usage stay the same. A model that uses fewer output tokens can reduce cost per completed task by more.

GPT API prices before and after the discount

The five-model prices below are the October 22 snapshot in USD per 1 million tokens, for prompts of at most 272K input tokens. "OpenAI list" means standard processing, not Batch, Flex, Priority, regional processing, or a negotiated enterprise contract. OpenAI figures are from its pricing page and reAPI figures from its live model pages, both checked October 22, 2026.^[3][5]

OpenAI publishes a separate cached-input price for these models. reAPI's model pages list short-context cache-read and cache-write rates for the three GPT-5.6 tiers (Sol: $0.32 per million cached input tokens and $4 per million cache-write tokens). These equal 80% of OpenAI's corresponding read and 30-minute write rates, but reAPI does not label the write lifetime; the prices alone do not establish matching retention controls.^[8][9] The GPT-5.4 and GPT-5.5 pages list no cached-input rate, so this comparison does not invent one. For those two, treat repeated prompt prefixes as a separate benchmark.

What the GPT API price cut means for a real bill

Suppose an agent processes 100 million uncached input tokens and generates 20 million output tokens each month, with each request staying within the short-context tier. With identical token usage and no tool charges, the monthly calculation is:

monthly cost = input MTok × input rate + output MTok × output rate

Historical illustration: this graphic predates the October 22 price update. Use the dated table below for the revised calculation; the amounts and blanket 20% discount in the graphic are not current quotes.

For the four models at 20% off, the percentage is constant and the dollar impact grows with spend. A team currently paying $10,000 per month for the same mix of uncached standard traffic would retain roughly $2,000 after moving it, or roughly $6,000 on GPT-5.4, before considering any change in model behavior or token consumption.

That last condition matters. API cost is not merely price per token:

cost per successful task
= price per token × tokens per attempt × attempts per success

A cheaper model that needs two retries can cost more than an expensive model that succeeds once. A more concise model can beat the table by emitting fewer billable output tokens. OpenAI says GPT-5.6 improves token efficiency, so a carefully evaluated migration may save more than the headline 20% per-token discount.^[1]

Which discounted GPT model should you use?

When this article first ran, the table had two same-price pairs. The current rates break both: on reAPI the older GPT-5.4 is now cheaper than Terra, and the newer Sol is cheaper than GPT-5.5.

GPT-5.6 Luna: the volume default

Luna is the lowest-priced model in this group. OpenAI positions it for cost-sensitive, high-volume work, with a 1.05M-token context window and 128K maximum output. It is the first candidate for classification, extraction, ranking, routine support, and inexpensive agent subtasks.^[4]

At $0.16 input and $0.96 output per million tokens on reAPI, Luna is 90% cheaper than Terra on both dimensions. Use the price gap to run a proper eval, not to skip one: measure task completion, retries, latency, and total output tokens.

GPT-5.6 Terra versus GPT-5.4: newer tier, higher reAPI price

On reAPI, GPT-5.4 costs $1 input and $6 output per million tokens and Terra costs $1.60 and $9.60, so Terra is 60% more per token. At OpenAI's list prices the order is reversed: Terra's $2/$12 is below GPT-5.4's $2.50/$15.^[5] Terra is still the natural first test for new work because OpenAI describes it as the balanced GPT-5.6 tier, competitive with GPT-5.5 while priced below the flagship.^[1]

GPT-5.4 is the cheaper option on reAPI, and it remains useful when an existing production prompt has already been validated against it. Moving to Terra costs more per token and is not automatically better: model behavior, reasoning defaults, cache behavior, and output length can change. Run a shadow comparison before moving all traffic.

GPT-5.6 Sol versus GPT-5.5: newer model, lower rate

On reAPI, Sol costs $3.20 input and $16 output per million tokens against GPT-5.5's $4 and $24: 20% less on input and a third less on output. OpenAI's list shows the same gap, $4/$20 against $5/$30, with Sol's rate labelled promotional through at least November 21, 2026.^[5] Sol is the new flagship and adds the GPT-5.6 generation's updated reasoning, tool use, and explicit prompt-caching controls, while GPT-5.5 is the stable baseline for teams that have already tuned around it.^[1][2]

For new high-complexity deployments, start the evaluation with Sol. Keep GPT-5.5 only when regression tests show that its established behavior wins on your workload.

When 20% off is not the cheapest option

A clean price table is useful, but three billing details can reverse a narrow comparison.

Batch and Flex processing

OpenAI offers Batch and Flex processing at 50% of standard rates for GPT-5.4, GPT-5.5, and the GPT-5.6 tiers.^[2][5] Batch is asynchronous, with a 24-hour completion window; Flex has slower responses and possible resource unavailability. Compare these modes with the same workload requirements rather than treating either as an identical real-time service.^[7]

At 50% off, OpenAI Batch beats a 20% discount. GPT-5.4 is the exception: reAPI's $1/$6 is below OpenAI's Batch rate of $1.25/$7.50.^[5] Otherwise the reAPI discount is most directly relevant to synchronous, standard API traffic. A nightly offline enrichment job has different economics from a live coding agent.

Prompt caching

OpenAI lists cached reads for GPT-5.6 at 10% of uncached input price. GPT-5.6 bills default 30-minute cache writes at 1.25 times the uncached input rate and 12-hour writes at 2.5 times. These replace ordinary input charges for the written tokens; they are not added on top. The net result depends on how often the prefix is reused.^[9][1]

For a long system prompt reused hundreds of times, cache hit rate can matter more than a 20% reduction in ordinary input price. Track uncached input, cache-write tokens, cache-read tokens, and output separately.

Long-context requests

For GPT-5.4 and GPT-5.6 models with 1.05M context windows, OpenAI prices prompts above 272K input tokens at 2x input and 1.5x output for the full request.^[4][6] reAPI's model pages list long-context rates for the GPT-5.6 tiers (Sol: $6.40 input and $24 output per million tokens). Terra's long-context output was displayed at $14.50, versus OpenAI's $18, so it is not exactly a 20% reduction. The October 22 comparison did not list reAPI long-context rates for GPT-5.4 or GPT-5.5. If your prompts cross the threshold, verify the applicable rate and billed usage before projecting savings.^[8]

How to use the lower GPT API rates

The GPT-5.6 tiers use reAPI's unified Chat Completions endpoint with a workspace API key. This example uses GPT-5.6 Terra; keep the SDK's Chat Completions call shape and configure its base URL:^[11]

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_REAPI_KEY",
    base_url="https://reapi.ai/api/v1",
)

response = client.chat.completions.create(
    model="gpt-5.6-terra",
    messages=[
        {"role": "user", "content": "Review this pull request for data-loss risks."}
    ],
)

print(response.choices[0].message.content)

The model ids in the five-model comparison are:

• gpt-5.4
• gpt-5.5
• gpt-5.6-sol
• gpt-5.6-terra
• gpt-5.6-luna

GPT-5.6 Sol, Terra, and Luna use the unified base shown above. GPT-5.4 and GPT-5.5 retain the legacy https://api.reapi.ai/v1 base and a separate gateway key and balance; follow their GPT-5.4 documentation and GPT-5.5 documentation. Keys from the two workspaces are not interchangeable.^[11]

Do not replace the dots with hyphens in API requests. The reAPI website uses hyphenated page slugs, but the wire model ids preserve OpenAI's dotted version numbers.

A safe migration checklist

Moving an API route is simple; proving that the new route is economical takes a little more work.

1. Export a representative sample. Include short and long prompts, tool calls, structured output, and known failure cases.
2. Pin the model id. Do not compare moving aliases if you need a stable before-and-after result.
3. Run both routes. Measure answer quality, first-token latency, total latency, input tokens, output tokens, retries, and errors.
4. Calculate cost per accepted result. Token price alone hides retries and human review.
5. Inspect feature compatibility. Validate tools, structured outputs, streaming, and image input used by the application.
6. Move traffic gradually. Start with a small percentage and keep the old route available until error rates and bills match the test.
7. Check the live rate card. Pricing can change after an article is published; treat the model page and API rate card as the source of truth.

FAQ

Did OpenAI lower GPT API prices?

For the GPT-5.6 family, yes. OpenAI's pricing page, checked October 22, 2026, lists GPT-5.6 Sol at $4/$20 (labelled promotional through at least November 21, 2026), Terra at $2/$12, and Luna at $0.20/$1.20 per million input and output tokens, all lower than when this article first ran. GPT-5.4 ($2.50/$15) and GPT-5.5 ($5/$30) are unchanged. The reductions described here are the reAPI rates relative to those current standard prices.^[5]

Which GPT models are 20% cheaper on reAPI?

GPT-5.5, GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna are 20% below OpenAI's standard short-context input and output rates in the October 22 comparison. GPT-5.4 is 60% below. This percentage does not cover every long-context or cache category.^[3]

What is the cheapest GPT-5.6 API model?

GPT-5.6 Luna. It costs $0.20 input and $1.20 output per million tokens at OpenAI's standard rate, or $0.16 input and $0.96 output on reAPI's current default route.^[3][4]

Is GPT-5.6 Terra cheaper than GPT-5.4?

Not on reAPI, where GPT-5.4 is $1/$6 and Terra is $1.60/$9.60. At OpenAI's list prices it is the other way round: Terra $2/$12, GPT-5.4 $2.50/$15. Terra is newer, but the best value depends on task success, token efficiency, and migration risk.

Is GPT-5.6 Sol cheaper than GPT-5.5?

Yes. On reAPI Sol is $3.20/$16 per million input/output tokens against GPT-5.5's $4/$24, and at OpenAI's standard rate it is $4/$20 against $5/$30. Sol offers the newer capability tier at the lower per-token price.^[1]

Does reAPI offer discounted cached input?

For the GPT-5.6 tiers, yes: the model pages list short-context cache reads at $0.32 (Sol), $0.16 (Terra), and $0.016 (Luna) per million tokens, 80% of OpenAI's corresponding cached-input rates. GPT-5.4 and GPT-5.5 have no published cached-input rate on reAPI, so do not assume one. Compare a real cached workload and check live billing before migration.

Can I keep using the OpenAI Python SDK?

Yes, for supported Chat Completions requests. For GPT-5.6, set base_url to https://reapi.ai/api/v1, use a unified-workspace reAPI key, and keep the canonical dotted, suffixed model id. GPT-5.4 and GPT-5.5 use the separate legacy endpoint and key described above. This does not make Responses requests interchangeable with Chat Completions.^[11]

The price cut is useful; cost per result is the decision

A 20% reduction is unusually easy to budget. Keep the workload and model fixed, and a $1,000 standard-token bill becomes roughly $800, or about $400 on GPT-5.4. The more valuable opportunity is model routing: Luna for routine volume, Terra for the balanced default, and Sol only where frontier capability improves the success rate enough to pay for itself.

The honest headline is therefore narrower than "OpenAI cut prices" and more useful to developers: the five models in the October 22 comparison run below OpenAI's standard short-context input/output rates on reAPI, four at 80% and GPT-5.4 at 40%. Verify your cache pattern, long-context usage, and asynchronous options, then measure cost per accepted result before moving production traffic.

References

1. OpenAI. GPT-5.6: Frontier intelligence that scales with your ambition. Pricing and tier details. openai.com/index/gpt-5-6

2. OpenAI. Introducing GPT-5.5. API pricing and processing options. openai.com/index/introducing-gpt-5-5

3. reAPI. GPT model pages and live rates. Retrieved October 22, 2026. reapi.ai/models

4. OpenAI. GPT-5.6 Luna model. Specifications, pricing, caching, and long-context rules. developers.openai.com/api/docs/models/gpt-5.6-luna

5. OpenAI. API pricing. Standard, Batch, Flex, and Fast processing rates. Retrieved October 22, 2026. developers.openai.com/api/docs/pricing

6. OpenAI. GPT-5.4 model. Pricing and long-context rules. developers.openai.com/api/docs/models/gpt-5.4

7. OpenAI. Current API prices and processing conditions. Checked October 23, 2026. API pricing, Batch, Flex.

8. reAPI. Rendered model pricing. Checked October 23, 2026. GPT-6 Astra, GPT-6 Sol, GPT-6 Luna, GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna.

9. OpenAI. Prompt caching and reasoning token accounting. Checked October 23, 2026. Prompt caching, Reasoning.

10. OpenAI. Managing billing for ChatGPT and the API platform. Checked October 23, 2026. Separate billing systems.

11. reAPI. Model-specific API endpoints and authentication. Checked October 23, 2026. GPT-5.6 Terra, GPT-5.4, GPT-5.5, GPT-6 Astra.

Further reading

• reAPI. How to use GPT-5.6. reapi.ai/blog/how-to-use-gpt-5-6
• reAPI. How to use GPT-5.5. reapi.ai/blog/how-to-use-gpt-5-5
• reAPI. GPT-5.6 model catalog. reapi.ai/models