Codex Service Pricing

Long context (input tokens > 272K)

gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna always use the standard rates above, with no long-context surcharge. Only models listed below use long-context rates when a request's input tokens strictly exceed 272K; output tokens do not count toward this threshold. The table below is 10% of OpenAI's published long-context standard API prices. After top-ups of $10 or more receive 3x credited balance, effective cash cost remains about 1 / 3 of the standard prices below.

OpenAI's published Fast/Priority price table currently lists short-context prices only. When upstream returns the long-context usage above, the system strictly bills at 10% of the published long-context standard price and does not guess an unpublished Fast multiplier.

GPT-5.3-Codex-Spark Temporary Promo Rate

gpt-5.3-codex-spark is now available on PhotonMark Codex. According to OpenAI's GPT-5.3-Codex-Spark introduction page, this is a Codex research preview model for fast coding workflows. It fits small code edits, local refactors, copy/config changes, short scripts, and UI polishing tasks that need quick feedback.

From July 26, 2026 at 17:15 NZST, PhotonMark's temporary rates are input USD $0.05 / 1M, cached input USD $0.005 / 1M, and output USD $0.30 / 1M. With 3x credited balance on USD $10+ top-ups, effective cash cost is about input USD $0.0167 / 1M, cached input USD $0.0017 / 1M, and output USD $0.1 / 1M. Historical usage keeps its original charge.

For long autonomous debugging, large refactors, complex reasoning, or more careful code review, we still recommend gpt-5.4, gpt-5.5 or a higher-tier model.

Default Billing For Unlisted Models

If the Codex client or OpenAI upstream returns a model not listed in this table, the system keeps the original model name for audit and charges it at the same rate as gpt-5.5-pro until we add that model to the public billing table based on official OpenAI prices and multipliers. This avoids undercharging new or unknown models.

Fast Mode Billing

Fast mode does not multiply upstream token usage and does not change the real input, cached input, or output token counts shown in the dashboard. It only means the request uses a faster service tier and is billed with a higher credit multiplier. The currently supported Fast models—gpt-5.6-sol, gpt-5.6-terra, gpt-5.6-luna, gpt-5.4 and gpt-5.5—are all calculated at 2.5x the standard cost; gpt-5.6 does not use a higher base rate because of context length. Other models receive a multiplier only after OpenAI publishes the corresponding Fast/Priority price. After OpenAI releases a new Codex model, we verify official prices and multipliers before adding it to the public billing table.

Realtime voice beta pricing

The experimental Codex Realtime V3 voice workflow is available to authenticated Pay users. A successfully connected Realtime sideband is deducted from the Pay balance at the standard rate of USD $0.01 per connected minute, settled by actual duration to the second. The first 10 seconds of each session are free and each session is limited to 10 minutes.

Failed call creation, HTTP 401/502 errors, or attempts that never establish the Realtime sideband incur no Realtime session charge. Historical tests before this pricing launch are not charged retroactively. Ordinary Codex model work triggered during a voice conversation remains subject to the applicable model and tool rates; the Realtime connection fee is not a replacement for those charges.

This is a usable beta based on the current gpt-live-1-boulder-alpha route, not a promise that the alpha model name or interface will remain permanently available. PhotonMark may adjust, pause, replace, or withdraw the beta if upstream compatibility, stability, safety, or cost changes. See the setup and testing notes.

Web Search Billing

If a request uses the OpenAI web search tool, web search calls are billed in addition to model input, cached input, and output tokens. The current rate is 10% of OpenAI's current web search price, USD $0.0010 per web search call, deducted from Pay / Boost balance. Tokens from search results are still billed as input tokens for the selected model.

Image generation pricing

Image generation and editing are billed per returned image according to model, quality, size orientation, and n. The current gpt-image-2 deductions below are 10% of OpenAI's published image-generation prices. Any main-model tokens used by a tool workflow are billed separately.

n multiplies the per-image deduction. Custom square sizes use the square billing class; wider sizes, including 3840x2160, use landscape pricing; taller sizes, including 2160x3840, use portrait pricing. This classification determines billing only and does not rewrite the requested dimensions.

If the image model, quality, or size is missing, auto, or not recognized for billing, the fallback is gpt-image-2, medium, 1024x1024: USD $0.0053 per image. See the Image API parameter and 4K notes.

Real Billing Example

If using gpt-5.4-mini, the price is input USD $0.075 / 1M, cached input USD $0.0075 / 1M, output USD $0.45 / 1M. If a response has input 20,000, cached input 10,000, and output 500, normal input is 10,000; the cost is USD $0.00075 + USD $0.000075 + USD $0.000225 = USD $0.00105, rounded up to USD $0.0011 at 4 decimal places.

If the same usage comes from gpt-5.5 Fast request, first calculate standard cost using gpt-5.5 input, cached input, and output rates, then multiply by the 2.5x Fast multiplier. Token counts still show real usage and are not changed to 2.5x because of Fast.

Note: dashboard balance deductions use the standard prices on this page. After USD $10+ top-ups receive 3x credited balance, effective cash cost is about 1 / 3 of the standard deduction.