OpenAI Cuts GPT-5.6 Luna and Terra Prices, Renames Priority Processing to Fast Mode

OpenAI announced a set of price and speed changes for the GPT-5.6 family: GPT-5.6 Luna is 80% cheaper, GPT-5.6 Terra is 20% cheaper, and Priority processing was renamed Fast mode. (OpenAI, OpenAI Fast mode docs)

OpenAI also said it is upgrading auto-review in the ChatGPT app and Codex CLI from GPT-5.4 to GPT-5.6 Luna, and that combined with Luna's new price it expects auto-review to cost about 10x less. (OpenAIDevs on X, OpenAI)

Update: Fast Mode Is Not Sol-Only (August 11, 2026)

This page originally described the July 30 change as introducing "Fast mode for GPT-5.6 Sol." Re-verified against OpenAI's own documentation, that framing was too narrow in one direction and too broad in the other, so both halves are corrected here.

OpenAI's Fast mode guide states: "Priority processing was renamed Fast mode on July 30, 2026. We also increased the speed at which Fast mode operates for gpt-5.6-sol to make it up to 2.5x faster than Standard processing." (OpenAI Fast mode docs)

So two separate things happened on the same day:

• The rename applies across supported models, not just Sol. OpenAI's pricing page publishes Fast mode rates for gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, and gpt-5.3-codex. (OpenAI pricing)
• The speed increase was specific to Sol. The "up to 2.5x faster" figure is what OpenAI attaches to gpt-5.6-sol in that sentence. (OpenAI Fast mode docs)

The original July 30 announcement post is not readable by automated verification (see Sources), so nothing on this page is restated as freshly confirmed from it. Everything below is checked against OpenAI's pricing page and Fast mode guide as of August 11, 2026.

Key takeaways

• Lower prices for Luna and Terra: OpenAI says Luna is 80% cheaper and Terra is 20% cheaper. (OpenAI)
• Fast mode is the new name for Priority processing, and it covers Sol, Terra, Luna, and gpt-5.3-codex -- not Sol alone. (OpenAI pricing)
• Fast mode costs exactly 2x Standard on every GPT-5.6 model OpenAI publishes a rate for. (OpenAI pricing)
• Two parameter values work: service_tier: "fast" and service_tier: "priority" both reach Fast mode on supported models. (OpenAI Fast mode docs)
• GPT-5.6 pricing has short-context and long-context rates, and the long-context rate is roughly double. Budgeting from the short-context number alone will understate long-context calls. (OpenAI pricing)
• ChatGPT/Codex subscriptions unchanged: OpenAI says subscription prices and quota budgets remain unchanged; Terra and Luna usage now consumes fewer credits. (OpenAI)

What changed (with numbers)

OpenAI framed this release as pushing the "price-performance frontier" by improving efficiency and passing those gains through to developers and teams.

API pricing

Standard rates, per 1M tokens, from OpenAI's pricing page (re-verified August 11, 2026). GPT-5.6 models are priced separately for short and long context:

Source for the whole table: OpenAI pricing. Sol pricing is unchanged by this release. (OpenAI)

Fast mode pricing

Fast mode is billed at exactly 2x the Standard rate on every model OpenAI publishes a Fast rate for:

Source: OpenAI pricing. Cached-input discounts still apply to Fast mode requests. (OpenAI Fast mode docs)

Turning Fast mode on

Per OpenAI's Fast mode guide:

• Per request: set service_tier to fast on a Responses or Chat Completions call. Setting it to priority "provides the same behavior for supported models."
• Per project: open Settings, select General under Project, and set Project Service Tier to Fast. Requests that omit service_tier then default to Fast mode, and the project transitions gradually rather than instantly.
• What the response reports: for GPT-5.6 and earlier models, the response object returns priority whether you sent priority or fast. The same applies to the usage dashboard when grouping by service tier.

(OpenAI Fast mode docs)

Two Fast mode limits worth knowing before you switch

Neither of these is visible from the pricing table, and both change what you actually get:

• The ramp rate limit can silently downgrade you. OpenAI says that if traffic ramps too fast, the system "may downgrade some Fast mode requests to standard speeds and charge standard rates," and the response then contains service_tier: "default". It may apply if you send at least 1 million tokens per minute and increase that by more than 50% within 15 minutes. OpenAI's advice is to ramp over hours, use feature flags to shift traffic, and keep large batch or ETL jobs off Fast mode. The limit is shared across all your projects and organizations.
• Fast mode shares your Standard rate limit. For a given model, Standard processing and Fast mode draw on the same rate limit, so paying the premium buys speed, not headroom. Fast mode also does not support fine-tuned models or embeddings, and it is separate from Scale Tier: Fast mode requests are billed separately and do not count against purchased Scale Tier token bundles.

(OpenAI Fast mode docs)

Auto-review (ChatGPT app + Codex CLI)

OpenAI says it is upgrading auto-review in the ChatGPT app and Codex CLI from GPT-5.4 to GPT-5.6 Luna, and expects the combination of the model switch plus Luna's new pricing to make auto-review about 10x cheaper. (OpenAIDevs on X)

This is the one claim on the page with no durable, re-checkable OpenAI-owned source: it was announced on X and in the July 30 post, and neither surface can be re-read by an automated check. It is reported here as announced on July 30, 2026, not as independently confirmed since.

What this means in practice

• If you run high-volume workflows, Luna's new pricing is the headline change, and Luna's Fast mode rate of $0.40 / $2.40 is still below Terra's Standard rate.
• If you want faster responses on latency-sensitive paths, Fast mode is available across the GPT-5.6 family at 2x Standard -- but the 2.5x speed figure OpenAI publishes is the one it attaches to Sol.
• If your traffic is spiky, read the ramp rate limit before budgeting: a request that gets downgraded returns service_tier: "default" and bills at Standard rates, so the premium is not a latency guarantee.
• If you use auto-review in ChatGPT or Codex CLI, OpenAI signaled a substantial expected cost reduction, but the only specific figure shared is the "about 10x cheaper" expectation.

Sources

Re-verified August 11, 2026:

• OpenAI, Pricing -- https://developers.openai.com/api/docs/pricing (Standard, Batch, Flex, and Fast rate tables; the Priority-to-Fast rename note)
• OpenAI, Fast mode guide -- https://developers.openai.com/api/docs/guides/fast-mode (rename and Sol speed increase, service_tier configuration, ramp rate limit, rate-limit sharing, Scale Tier separation)

Cited as published on July 30, 2026, and not re-verifiable by automated checks:

• OpenAI: Advancing the price-performance frontier with GPT-5.6 -- https://openai.com/index/advancing-the-price-performance-frontier-with-gpt-5-6/ (the whole openai.com HTML surface returns HTTP 403 to automated fetches; readable in a browser)
• OpenAI post on X (Jul 30, 2026) -- https://x.com/OpenAI/status/2082878156483219672 (x.com returns HTTP 402 to automated fetches)
• OpenAI Developers post on X (Jul 30, 2026) -- https://x.com/OpenAIDevs/status/2082878497043923265 (same)