OpenAI's cheap tier becomes a routing problem: Terra $2/$12, Luna $0.20/$1.20, seats untouched
The July 30 cuts move the argument from model access to per-step token cost. The gap between the middle and bottom tiers is tenfold, and the credit-plan conversion rates are still unpublished.
The Engineer · Build desk
What happened
- OpenAI's lower GPT-5.6 pricing for its Terra and Luna models took effect July 30 through the API.
- OpenAI characterised the two changes as reductions of 20% for Terra and 80% for Luna.
- Included usage on Pro, Plus and Business subscriptions was left as it was.
- Lower credit consumption is rolling out on eligible Work and Codex plans, but the announcement names neither the qualifying plans nor their new conversion rates.
Why it matters
- Who has to decide decision The document that sets an agent budget is now a routing table rather than a price list, because whether a given step lands on Terra or Luna is a tenfold difference in that step's cost.
- What it limits constraint Cheaper tokens shrink the size of a bad month without making it forecastable, so the job of predicting what a long-running agent will spend stays with the buyer.
- Who's exposed exposure Teams inside credit plans are approving a discount they cannot verify, and their finance function has nothing to reconcile against until the conversion rates appear.
- What it sets up precedent With seat allowances untouched, a competitor's answer now has to arrive as a cheaper low tier rather than a cheaper subscription, which is the fight runtimewire expects next.
Next story · Build · Oct 8
OpenAI sells the same GPT-6.1 Sol model faster for six times the token priceConfirmed2 publishers
Run the percentages backwards and the old price card appears. A 20% reduction that lands on $2 and $12 implies Terra was $2.50 and $15 [16]. An 80% reduction that lands on $0.20 and $1.20 implies Luna was $1 and $6 [17]. OpenAI published one percentage per model [1], so that reconstruction assumes the same cut applied to input and output. If it did not, the input side moved by something other than the published figure, and that is the first thing to check against a July invoice.
The more useful number is the ratio inside each card. Terra and Luna both bill output at six times input [18], which means output becomes the larger half of the bill once it reaches roughly 14% of the tokens in a run [19]. Agent work sits well above that threshold. ChatGPT Work and Codex search through context, call tools, write files, run tests and revise their output, and every one of those steps consumes tokens [12]. The rate that governs a long agent run is the output rate, not the input rate the headline leads with.
Which is why the line that matters runs between tiers, not between last month and this one. Terra costs ten times Luna on both input and output [20]. OpenAI says Luna can use tools and complete multi-step workflows [6], and pointed to document analysis, classification and routine implementation as the work that becomes economical at greater volume [9]. Nothing in the pricing announcement specifies an architecture; the pattern of reserving the top tier for planning and handing implementation or testing to a cheaper one is runtimewire's inference from the tier descriptions [2]. It is also the only way to collect the tenfold gap, and it is a classifier plus a step budget that somebody on staff writes and then maintains.
The two headline rates are not the whole card either. The API documentation lists separate rates for cached input, long-context requests and other processing modes [13], and a multi-hour run touches several of them.
For subscription customers the arithmetic is not yet available at all. Pro, Plus and Business usage is unchanged [11], while eligible Work and Codex plans are getting lower credit consumption for Terra and Luna [5] with no per-plan conversion rate published in the announcement [8]. API buyers can re-model this week. Credit-plan buyers have been handed an improvement they cannot size until they inspect their own workspace configuration.
Either way the discount only shows up for teams that already count tokens per step. A workflow whose output share per run is unknown cannot be tested against these numbers, and its owner has no way to say whether 20%, 80% or nothing arrived [19].
What to watch
- Publication of per-plan credit conversion rates for ChatGPT Work and Codex, which is what would let subscription customers price the change they received.
- Any clarification from OpenAI on whether the 20% and 80% figures applied to input and output tokens alike, which decides what the prior card looked like.
- Whether Luna actually finishes tool-calling steps in production, since the tenfold saving only lands if the cheap tier does not need a retry on the expensive one.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence62
- Adoption
- Not scored
- Hype gap+15
- Incentives60
- Confidence63
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
OpenAI described the Terra and Luna changes as reductions of 20% and 80% respectively.
- [2]
The routing pattern of reserving Sol for planning and handing implementation, classification or testing to a cheaper tier is an inference by runtimewire from the capabilities and price tiers OpenAI described, not an architecture specified in the pricing announcement.
- [3]
OpenAI said lower GPT-5.6 pricing for its Terra and Luna models is available through the API and rolling out across eligible ChatGPT Work and Codex credit plans, with the pricing and credit changes taking effect July 30.
- [4]
Long-running agents can consume widely different amounts of context and output even when they begin with similar instructions; lower Terra and Luna rates reduce that exposure while the underlying variability remains.
- [5]
OpenAI's post says eligible ChatGPT Work and Codex plans are rolling out lower credit consumption for Terra and Luna.
- [6]
OpenAI said Luna can use tools and complete multi-step workflows, positioning the least expensive GPT-5.6 model as an agent component rather than a basic text generator.
- [7]
OpenAI launched the GPT-5.6 family on July 9 in three tiers: Sol for complex reasoning and coding, Terra balancing capability and cost, and Luna for high-volume work where price matters most.
- [8]
OpenAI has not publicly specified in the supplied announcement which plans qualify for the credit change or provided each plan's new conversion rate.
- [9]
The July 30 price-performance announcement emphasized document analysis, classification and routine implementation as workloads that could become economical at greater volume.
- [10]
API customers pay $2 per million input tokens and $12 per million output tokens for GPT-5.6 Terra, and $0.20 per million input tokens and $1.20 per million output tokens for GPT-5.6 Luna.
- [11]
Pro, Plus and Business subscription usage remains unchanged, according to OpenAI's post.
- [12]
ChatGPT Work and Codex can search through context, call tools, write files, run tests and revise their output, and each step consumes tokens.
- [13]
The API pricing documentation lists separate rates for cached input, long-context requests and other processing modes.
- [14]
Altman wrote in June that advanced AI needs to become abundant, affordable and practical enough for broad use.
- [15]
OpenAI introduced ChatGPT Work on July 9 as an agent for longer projects across connected apps and files, able to create documents, spreadsheets, presentations and websites and to continue projects through scheduled tasks.
- [16]
If the stated 20% reduction applied equally to both sides of Terra's card, its prior rates were $2.50 per million input tokens and $15 per million output tokens.
- [17]
If the stated 80% reduction applied equally to both sides of Luna's card, its prior rates were $1.00 per million input tokens and $6.00 per million output tokens.
- [18]
Both Terra and Luna bill output tokens at six times the input rate.
- [19]
At a 6:1 output-to-input rate, output costs exceed input costs once output is more than about 14.3% of the tokens in a run.
- [20]
Terra costs exactly ten times Luna on both input and output tokens.
- [21]
runtimewire.com argues that lowering the cost of long-running agent work without cutting subscription prices makes repeatable workflow economics, rather than model access alone, the next battleground.
Sources
2 independent publishers whose own reporting we read for this story.
- dev.toOpenAI GPT-5.6 Pricing Update Cuts Terra and Luna Costs, Leaves Sol Unchanged
2 articles · August 22, 2026
- runtimewire.comOpenAI cut GPT-5.6 API prices and agent credit use on July 30
1 article · August 21, 2026