By SaaS Feeds Editorial
OpenAI's published API prices for the six current-generation GPT-6 and GPT-5.6 models run from $0.10 per 1 million input tokens on GPT-6 Luna to $10.00 per 1 million input tokens and $50.00 per 1 million output tokens on GPT-6 Astra, with input and output billed separately and every figure in US dollars [3]. The changelog's active list also carries seventeen older models with published API rates, from GPT-4o Mini at $0.15 in and $0.60 out per 1M tokens up to GPT-5.5 Pro at $30.00 in and $180.00 out, plus GPT-5.5 Instant listed as a ChatGPT-side model without API pricing; Table 2 lists all eighteen [3]. The only rate on the ladder with an announced floor is GPT-5.6 Sol, cut on August 21 to $4.00 per 1M input and $20.00 per 1M output tokens as a promotional price "for three months, through at least November 21, 2026" [3], which puts the floor two days after this article's date.
The September launches carry the rates to plan against. GPT-6 Sol's $2.00 input and $10.00 output are exactly half of GPT-5.6 Sol's $4.00/$20.00, and GPT-6 Luna's $0.10 input is half of GPT-5.6 Luna's $0.20, with Luna's output at $0.50 against the predecessor's $1.20 [3][6][8]. OpenAI says, as recorded in the changelog, that the Sol cut is permanent rather than promotional [3][6]. OpenAI's pricing pages returned HTTP 403 during reporting, so each rate in this article is attributed to the fetched source named with it; the official pricing page is the final check before committing spend [3].
Table 1. List prices for the six current-generation GPT-6 and GPT-5.6 models as of November 19, 2026, in USD per 1M tokens with input and output billed separately [3]. The GPT-6 rows are corroborated by the per-model pages [6][7][8]; the GPT-5.6 Terra and Luna rows also appear in the wire report and the developer post [1][5].
| Model | Input | Output | Status | Published conditions |
|---|---|---|---|---|
| GPT-6 Astra | $10.00 | $50.00 | limited preview since Sep 3 | cached input $1.00; Batch at half price; Fast mode at 2x; above about 272K input tokens billed $20.00/$75.00 |
| GPT-6 Sol | $2.00 | $10.00 | released Sep 22 | cached input $0.20 |
| GPT-6 Luna | $0.10 | $0.50 | released Sep 22 | cached input $0.01 |
| GPT-5.6 Sol | $4.00 | $20.00 | promotional, floor Nov 21, 2026 | announced Aug 21, more than 20% off |
| GPT-5.6 Terra | $2.00 | $12.00 | price in effect since Jul 30 | lower credit consumption in subscription entry points |
| GPT-5.6 Luna | $0.20 | $1.20 | price in effect since Jul 30 | lower credit consumption in subscription entry points |
Table 1 covers the rows the July 30, August 21, and September 2026 pricing moves touched; the entries in Table 2 stand at their current list prices [3].
Table 2. The eighteen remaining active models on the changelog's list, in USD per 1M tokens, in/out, with context windows [3].
| Model | Input | Output | Context |
|---|---|---|---|
| GPT-5.5 Pro | $30.00 | $180.00 | 1M |
| o1 | $15.00 | $60.00 | 200K |
| o3 | $10.00 | $40.00 | 200K |
| GPT-5.5 | $5.00 | $30.00 | 1M |
| GPT-5.4 | $3.00 | $12.00 | 1M |
| o1-mini | $3.00 | $12.00 | 128K |
| GPT-4o | $2.50 | $10.00 | 128K |
| GPT-5 | $2.00 | $8.00 | 128K |
| GPT-4.1 | $2.00 | $8.00 | 1M |
| GPT-5.2 | $1.50 | $6.00 | 512K |
| GPT-5.1 | $1.25 | $5.00 | 256K |
| o4-mini | $1.10 | $4.40 | 200K |
| o3-mini | $1.10 | $4.40 | 200K |
| GPT-5.4 Mini | $0.75 | $4.50 | 400K |
| GPT-4.1 Mini | $0.40 | $1.60 | 1M |
| GPT-5.4 Nano | $0.20 | $1.25 | 400K |
| GPT-4o Mini | $0.15 | $0.60 | 128K |
| GPT-5.5 Instant | N/A (ChatGPT-side) | N/A | 128K |
The September GPT-6 launches and the permanence claim on Sol
One of the two September launches carries OpenAI's commitment to last. GPT-6 Sol launched on September 22 at $2.00 per 1M input and $10.00 per 1M output tokens, half of GPT-5.6 Sol's then-current $4.00/$20.00, with cached input at $0.20 per 1M tokens, and the changelog entry records OpenAI's position that the cut is permanent rather than promotional [3][6]. GPT-6 Luna launched the same day at $0.10 in and $0.50 out per 1M tokens with cached input at $0.01, and the changelog describes the $0.10 input level as a new price floor for the low end of the lineup [3][8]. Both models ship with a 1,050,000-token context window, 128K max output, and knowledge cutoffs in April and May 2026 [3].
The Cailianshe wire's launch coverage assigns the flagship role on coding, biology, and cybersecurity tasks to Sol, everyday work to Terra, and fast, low-cost volume to Luna [1]. The changelog's GPT-6 descriptions continue that split, with Luna for high-volume chat, classification, and lightweight agentic tasks, Sol as the workhorse approaching Astra reliability at a fraction of the cost, and Astra as the frontier preview priced at 2.5 times the then-current GPT-5.6 Sol rate, with cyber-sensitive capabilities behind a trusted-access program [3][7].
The changelog still lists GPT-5.6 Sol and GPT-5.6 Luna as active alongside their successors [3]. That coexistence has a direct cost consequence. The older flagship GPT-5.6 Sol bills $4.00/$20.00 under its promotion while the successor bills $2.00/$10.00 with no announced end date, so the newer model is the cheaper line item on the same workload.
The July 30 entry-tier cuts and the August 21 Sol promo
The September resets extend a repricing pattern that began three weeks after the GPT-5.6 family shipped. On July 30 OpenAI cut GPT-5.6 Luna by 80% to $0.20 per 1M input and $1.20 per 1M output tokens and cut GPT-5.6 Terra by 20% to $2.00/$12.00, according to the Cailianshe wire report [1]. Sol's standard API rate, the ChatGPT and Codex subscription prices, and the quota budgets stayed unchanged, and Terra and Luna consume fewer credits in the subscription entry points where they run [1][5].
The wire's framing covered the announcement itself. Sam Altman presented a new pricing matrix for the GPT-5.6 line, with the entry-tier Luna cut as its headline number [2]. The recorded scope is narrower. Luna moved from $1.00/$6.00 to $0.20/$1.20 and Terra from $2.50/$15.00 to $2.00/$12.00, while GPT-5.6 Sol Standard held at $5.00/$30.00 [1][3][5]. Sol's July addition was a paid speed tier, a Fast mode announced at up to 2.5 times standard response speed for twice the standard price with the model's intelligence unchanged [2], and on the same day the API renamed priority processing to Fast mode as a service tier [3].
Table 3. GPT-5.6 list prices at launch on July 9 and after July 30, in USD per 1M tokens, in/out [3][5].
| Model | Launch price (in/out) | After July 30 (in/out) | Change |
|---|---|---|---|
| GPT-5.6 Luna | $1.00 / $6.00 | $0.20 / $1.20 | −80% |
| GPT-5.6 Terra | $2.50 / $15.00 | $2.00 / $12.00 | −20% |
| GPT-5.6 Sol Standard | $5.00 / $30.00 | $5.00 / $30.00 | unchanged |
Three weeks later the promotional line appeared. On August 21 OpenAI cut GPT-5.6 Sol by more than 20%, to $4.00 in and $20.00 out per 1M tokens for API and credit pricing, "for three months, through at least November 21, 2026" [3]. The wire context around these moves is a price competition the Cailianshe report ties to Moonshot AI's open-weight Kimi K3, Anthropic's Claude Opus 5 priced at half of Claude Fable 5, and Google's cost-focused Gemini releases [1]. Those figures are the wire's reporting; the counterpart rates were not independently fetched.
What the rate card leaves out
The input and output rates are the entry line on the bill. Four conditions and components from the fetched sources sit on top of them.
Cached input is the first discount worth checking, because production traffic rereads the same system prompts and context. The published cached rates are $0.20 per 1M tokens on GPT-6 Sol, $0.01 on GPT-6 Luna, and $1.00 on GPT-6 Astra [3][6][7][8]. No cached rate appears for the GPT-5.6 family in the fetched sources.
Astra's rate carries added conditions. The changelog lists Batch processing at half price, Fast mode at 2x, and a long-context surcharge that bills requests above about 272,000 input tokens at $20.00 in and $75.00 out [3][7]. Availability gates the price. Astra entered a limited preview on September 3 and reached paid ChatGPT users on September 4 in a restricted version that declines certain cybersecurity prompts [3][7].
Channel choice changes the number too. Amazon Bedrock carries the GPT-5.6 models and announced price cuts of up to 80% for them, per the AWS announcement headline, which is the only claim that page supports; no Bedrock-specific rate is citable from it [4]. On the resale side, a third-party gateway catalogue listed GPT-6 Sol at $1.30 in and $6.50 out per 1M tokens in its October 10 snapshot, below the $2.00/$10.00 list. That is a dated gateway figure, and it is not OpenAI pricing [6].
The line that absorbed most of an 80% token cut in one worked example is everything around the tokens. The developer post's formula prices each successful task as token cost plus retries plus tools and retrieval plus latency plus error cost [5]. Its hypothetical, labeled as such by the author and recomputable from the stated assumptions, takes a Luna attempt at 50,000 input tokens and 5,000 output tokens, a $0.01 tool and retrieval cost, a $0.004 latency cost, an 80% success rate implying 1.25 attempts per success, and a $0.02 human review and error reserve per success [5].
Table 4. The worked hypothetical from the developer post, per successful task, in US dollars [5].
| Component | at $1.00/$6.00 | at $0.20/$1.20 |
|---|---|---|
| First-attempt tokens | $0.0800 | $0.0160 |
| Expected retry tokens | $0.0200 | $0.0040 |
| Tools and retrieval | $0.0125 | $0.0125 |
| Latency | $0.0050 | $0.0050 |
| Review and error reserve | $0.0200 | $0.0200 |
| Total per successful task | $0.1375 | $0.0575 |
Under those assumptions the token price falls 80% and the cost per successful task falls about 58.2%. The example is a labeled hypothetical, not an average bill. The post notes that a higher cache-hit rate and a higher first-try success rate together push the realized saving toward the headline discount [5].
Three checks before the next budget cycle
Three checks put this list to work. First, confirm the current ladder on OpenAI's official pricing page [3]. Second, treat the $4.00/$20.00 GPT-5.6 Sol rate as ending at its announced floor. Volume that runs past November 21 prices out at $2.00/$10.00 on GPT-6 Sol unless the promotional rate is confirmed to continue [3]. Third, recompute cost per successful task with your own retry, tool, latency, and error numbers before assuming any token discount passes through, using the formula and the worked example as the template [5].
The fetched record leaves two items open, and each changes a budget differently. Whether the promotional GPT-5.6 Sol rate continues past its floor is not recorded; the announcement sets the floor and no end date [3]. Whether GPT-6 Astra has left its limited preview is likewise not recorded; the last dated entry is the September 4 restricted rollout to paid ChatGPT users [3][7].
When a token rate moves, the token lines of a real bill move with it and the tool, retrieval, latency, review, and error lines hold their dollar amounts. The worked example's 80% token discount therefore lands as a 58.2% saving per successful task; the saving approaches the headline discount only when the cache-hit rate and first-try success rise together, and the same arithmetic sets how much the cached, Batch, and Fast mode rates in Table 1 actually return [5].