OpenAI dropped the API prices on two of its three GPT-5.6 models on July 30, 2026, cutting the cheapest tier by 80% barely three weeks after the family reached general availability. The move signals how aggressively the leading model maker is competing for high-volume, cost-sensitive workloads — and how closely its customers are now counting every token. For more on the broader shift in AI economics, follow our latest AI developments.

What changed

According to OpenAI's published rate card and API changelog, the standard per-million-token prices now read:

  • GPT-5.6 Luna: 20 cents input and $1.20 output, down from $1 and $6 — an 80% reduction
  • GPT-5.6 Terra: $2 and $12, down from $2.50 and $15 — a 20% cut
  • GPT-5.6 Sol (the flagship): $5 and $30, unchanged

All three tiers launched to general availability on July 9, 2026 at the higher rates, placing the repricing just 21 days into the family's commercial life. Reuters and CNBC both confirmed the cuts, and Axios reported that the steepest reduction fell on the Luna model.

Cuts flow through every service tier

The reductions are not limited to headline pricing. They flow through every option on the rate card:

  • Batch and Flex processing, both half the standard price, put Luna at 10 cents input and 60 cents output per million tokens.
  • Cached input reads, discounted 90%, drop to two cents per million tokens on Luna and 20 cents on Terra.
  • Long-context requests, billed at double the input rate and 1.5 times output, land at 40 cents and $1.80 for Luna.

For teams already routing bulk classification, extraction, or long agent loops through the cheaper tiers, the same calls now cost a fraction of what they did a day earlier.

How the cuts compare to Anthropic

At 20 cents in and $1.20 out, Luna now undercuts Anthropic's cheapest published model, Haiku 4.5, by roughly five times on input and four times on output, according to Anthropic's pricing page. Terra's new rate sits below the $3 and $15 that Claude Sonnet 5 is scheduled to charge once its introductory pricing lapses on August 31, 2026.

At the top of both lineups, Sol still costs more on output than Anthropic's Opus 5, which is priced at $5 and $25.

Priority Processing becomes Fast mode

The same changelog entry retired "Priority Processing" and replaced it with "Fast mode." For the flagship Sol model, OpenAI says Fast mode runs up to 2.5 times standard speed at twice the price, and the switch is backward compatible — requests already tagged for priority route to Fast mode without a code change.

Fast-mode rates are $10 and $60 for Sol, $4 and $24 for Terra, and 40 cents and $2.40 for Luna. Notably, Anthropic sells the same product under the same name and the same terms, and the two rate cards now converge on region-pinned inference as well: OpenAI charges a 10% uplift on models released on or after March 5, 2026 when a customer requires data residency, while Anthropic bills US-only inference at 1.1 times its standard rate.

Why the cheaper tiers got cheaper

A day before the cut, OpenAI published an engineering post describing optimizations across its inference stack. Five members of the company's technical staff wrote that Sol, running inside its Codex coding tool, rewrote production GPU kernels — work OpenAI says cut end-to-end serving costs by 20%. The model also redesigned its own speculative-decoding draft model across hundreds of experiments, a change credited with raising token-generation efficiency by more than 15%.

Those are OpenAI's own figures for its own stack, but they point at the same cost center the price cut addresses: the agent harness behind Codex and ChatGPT. OpenAI caps tool output at 10,000 tokens by default and keeps model-visible history append-only, so an agent loop that resends its instructions at every step hits the prompt cache instead of paying full input rates. The company closed the post with a commitment to pass "under-the-hood improvements back to our users in the form of more widely available, cost-efficient intelligence." The rate card followed a day later.

Cloud partners and billing

OpenAI noted that buyers routing traffic through Amazon Bedrock are billed by AWS, and those rates can differ from its own published card. As more enterprises centralize model access through a single cloud vendor, the gap between OpenAI's direct pricing and what customers actually pay through intermediaries will matter as much as the headline numbers themselves.

A market counting every token

The cuts land as enterprise buyers audit inference spend more closely than ever. Earlier in the week, reporting documented how the era of unrestrained "tokenmaxxing" — pushing volume without watching cost — is fading as corporate AI bills spike across major platforms. OpenAI's decision to pass its internal cost savings down to customers, rather than bank the margin, is a clear signal of where it believes the competitive battle is now being fought: not at the frontier, where Sol still commands premium pricing, but on the cheap, high-volume tiers where most production traffic actually lives.

With Sol's price unchanged, the live decision for developers stays exactly where OpenAI has put it since the family shipped: which tier each request requires. The difference is that the cost of getting that decision wrong has just dropped dramatically.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →