Skip to content
TEN Brief Ten verified stories a day 2026.08.23 KO

이 기사는 한국어로도 읽을 수 있습니다 →

Tech · 3 min read · Breaking

GPT-5.6 Sol API price cut (August 21) — output falls from $30 to $20 per million tokens

On August 21, 2026 OpenAI cut API pricing for its frontier model GPT-5.6 Sol. Input fell from $5 to $4 per million tokens (-20%), output from $30 to $20 (-33%) and cached input from $0.50 to $0.40 (-20%). The rate is promotional, running through at least November 21, 2026, and is the company's second pricing move in under a month

A sunlit open-plan office in the morning with two laptops on a pale wooden desk, a plant and a coffee cup by the window

The three lines

  • Price — input $5→$4, output $30→$20, cached input $0.50→$0.40 per million tokens. The 33% cut to output is the largest
  • Duration — promotional, through at least November 21, 2026. This is not a permanent reduction
  • Scope — the API and eligible credit plans for ChatGPT Work and Codex. Pro, Plus and Business subscription prices are unchanged

Key questions

How much did GPT-5.6 Sol prices fall?
Per million tokens: input from $5 to $4 (-20%), output from $30 to $20 (-33%), cached input from $0.50 to $0.40 (-20%). OpenAI's phrase, 'more than 20%', describes the smallest of the three. The one that moves a developer bill most is the 33% on output. In most real workloads the token count is dominated by input — documents, code and conversation history all count as input — but output is priced five times higher, so it takes the larger share of the invoice.
How long does this price last?
OpenAI listed the rate as promotional and available through at least November 21, 2026 — a three-month window rather than a permanent cut. The distinction matters for costing. If you set a service price on the assumption of these rates, your input cost can rise on November 22 while your own price is already fixed. This page could not find an OpenAI statement on whether the rate reverts to the previous level after the window or moves to a different one.
Do subscription prices fall too?
No. What changed is API pricing and eligible credit plans for ChatGPT Work and Codex. Pro, Plus and Business subscriptions are unchanged. In other words the cut applies to calling the model from a program, not to the monthly fee a person pays to use the chat interface. Those are separate billing systems at OpenAI, and this announcement touched only the developer side.

AI model prices have moved in one direction for two years: down. Frontier models were the exception — the best model in a lineup rarely got cheaper.

On August 21, 2026 OpenAI broke that pattern.

1. What changed

ItemBeforeAfterChange
Input (per 1M tokens)$5$4-20%
Output (per 1M tokens)$30$20-33%
Cached input (per 1M tokens)$0.50$0.40-20%

OpenAI described the move as a cut of "more than 20%". That describes the smallest of the three.

The line that moves a bill most is output, and the reason is structural:

  • By token count, input usually dominates — documents, code and conversation history are all input.
  • By unit price, output is five times more expensive ($4 against $20).

So a summarisation workload, where input overwhelms output, saves close to 20%. A code-generation or long-answer workload saves more than 30%.

2. This price has a three-month clock

The rate carries a condition: it is promotional.

ItemDetail
TypePromotional rate, not a permanent cut
Minimum durationthrough November 21, 2026
After thatNot stated

The distinction has practical force. Set a service price against these rates and your input cost can rise on November 22 while your own price is already committed. Whether the rate reverts to the old level or moves somewhere else is not something this page could find OpenAI stating.

3. The second move in under a month

DateModelChange
July 30, 2026GPT-5.6 Luna$0.20 / $1.20 per 1M — roughly 80% lower
July 30, 2026GPT-5.6 Terra$2 / $12 per 1M — roughly 20% lower
August 21, 2026GPT-5.6 Sol$4 / $20 — 20–33% lower

Late July cut the cheaper tiers; August cut the frontier tier. The sequence ran from the bottom of the lineup upward.

Not everyone is moving the same way. This page covered the opposite case on August 18 in "DeepSeek raises API prices on August 16", where output pricing rose by as much as 371%. In the same market, one vendor is cutting and another is raising.

4. The gap to the nearest competitor

Several outlets read the cut as aimed at Anthropic's Claude Opus 5.

ModelInput (per 1M)Output (per 1M)
GPT-5.6 Sol (after cut)$4$20
Claude Opus 5$5$25

If both figures hold, GPT-5.6 Sol is 20% cheaper on both sides after the cut. The Anthropic figures here are quoted from press coverage; this page did not check them against Anthropic's own pricing page.

Unit price does not settle the choice on its own, because models differ in how many tokens they spend on the same job. OpenAI has claimed GPT-5.6 Sol reaches the same result with fewer tokens. If that holds, token consumption matters more to the final bill than the unit price does. This page could not find an independent measurement placing the two models' token usage side by side on identical tasks.

5. Why now

Two explanations are offered.

Competition. Anthropic reported second-quarter 2026 revenue of $11.6bn against OpenAI's $6.7bn, passing it for the first time (covered on this page on August 21), and its public IPO filing became known on August 21 (covered August 22). Chinese models are competing on price as well.

Cost. What it costs to serve a frontier model depends heavily on how GPU depreciation is booked, which changes reported profit without changing the hardware (this page, August 22, "What GPU depreciation is"). Where a company judges that a lower price is still profitable therefore varies. OpenAI has published no cost rationale for this cut.

6. What is not confirmed

  • After November 22, 2026 — no statement on the rate. The first thing to check if your costs are built on these numbers.
  • Eligible plans — which ChatGPT Work and Codex credit plans qualify was not confirmed.
  • Claude Opus 5 pricing — quoted from coverage, not checked at source.
  • Token consumption — no independent side-by-side measurement of tokens spent per task was found. Unit prices alone do not settle cost.
  • Next checkpoint — November 21, 2026, the end of the minimum promotional window.

Sources

  1. 93.3 The Drive — OpenAI cuts developer pricing for frontier GPT-5.6 Sol model by more than 20%
  2. GCC Business News — OpenAI cuts GPT-5.6 Sol API prices by more than 20%
  3. BigGo Finance — OpenAI Slashes GPT-5.6 Sol Developer Pricing by Over 20% to Counter Competition
  4. Startup Fortune — OpenAI Cuts GPT-5.6 Sol API Prices After Holding the Line for Months
  5. Storyboard18 — OpenAI slashes GPT-5.6 Sol API prices by more than 20% amid competition
  6. OpenAI — GPT-5.6: Frontier intelligence that scales with your ambition
  7. aipricing.guru — OpenAI API Pricing (August 2026)

Verification

Published
Last modified
Cross-check
Checked against 7 independent sources.
Unverified
  • This page could not find an OpenAI statement on what happens to the rate after the promotional window closes on November 22, 2026
  • The comparison figure of $5/$25 per million tokens for Anthropic's Claude Opus 5 is quoted by several outlets but was not checked against Anthropic's own pricing page. The comparison table rests on that assumption
  • Which specific 'eligible' ChatGPT Work and Codex credit plans are covered could not be confirmed
Authoring
Reviewed by a person before publication. The full process is described in the Editorial.

Ten stories, once each morning

We send the three-line summaries only; the full pieces stay on the site. One-click unsubscribe, any time.

Related