Capital & Compute
· Updated July 31, 2026· ai· pricing· economics

GPT-5.6 Pricing: What to Expect Per Task

GPT-5.6 launched June 26 2026 as Sol, Terra and Luna. The flagship Sol holds GPT-5.5 pricing at $5/$30 per million tokens. Here is the real cost per task.

By Capital & Compute

The original analysis below was written on June 19, before the launch. As of that date GPT-5.6 had no release: no announcement, no rate card, only a rumor with money behind it, with prediction-market traders putting the odds of a public launch by June 30 around 80 to 89 percent. The honest answer to “what does GPT-5.6 cost” was a projection off what the last two models actually cost. That projection has now resolved: the flagship Sol priced at exactly GPT-5.5’s $5 / $30, the “hold the line” case.

That starting point was unambiguous. GPT-5.5 runs $5 per million input tokens and $30 per million output tokens, and GPT-5.6 Sol landed on the same rate, with the real cost of using it decided less by the sticker rate than by how many tokens your task burns.

What GPT-5.5 costs right now

This is the one set of numbers here that is not a guess. GPT-5.5, released April 23, 2026, is the current OpenAI flagship, and its API pricing is public:

Tier Input / Mtok Output / Mtok Cached input / Mtok
GPT-5.5 standard $5.00 $30.00 $0.50
GPT-5.5 Pro $30.00 $180.00 varies
Long context (>272K) $10.00 $45.00 $1.00

The cached-input rate matters more than it looks. At $0.50 per million, repeated context (a system prompt, a codebase, a long document you keep referencing) costs a tenth of fresh input. Any honest GPT-5.6 cost estimate has to assume caching carries forward, because it has survived every recent OpenAI release.

The trajectory: GPT-5.4 to 5.5 doubled

Here is the pattern that should set your expectations. GPT-5.4, released March 5, 2026, was priced at $2.50 input and $15 output per million tokens, per model trackers like Vantage. Seven weeks later GPT-5.5 arrived at exactly double: $5 and $30, confirmed on OpenRouter’s model page and OpenAI’s own card.

OpenAI flagship token rates, GPT-5.4 to GPT-5.5Slope chart of per-million-token rates. Output rose from 15 dollars on GPT-5.4 to 30 dollars on GPT-5.5. Input rose from 2.50 dollars to 5 dollars. Both rates doubled across the two releases.GPT-5.4GPT-5.5Output$15.00$30.00Input$2.50$5.00
OpenAI flagship token rates, GPT-5.4 to GPT-5.5
ItemGPT-5.4GPT-5.5
Output$15.00$30.00
Input$2.50$5.00
Both the input and output rates doubled from GPT-5.4 (March 2026) to GPT-5.5 (April 2026). The open question is whether GPT-5.6 repeats the move, holds the line, or splits into tiers.Source: OpenAI API pricing; Vantage and OpenRouter model trackers

A single doubling is not a law. But it is the most recent data point, and it came with a capability jump that OpenAI clearly felt buyers would pay for. The question for anyone budgeting is whether 5.6 repeats that move, holds the line, or splits the difference with a cheaper fast tier.

What GPT-5.6 will likely cost: three scenarios

No GPT-5.6 number below is confirmed. Each is a scenario built off the GPT-5.5 baseline and the rumored feature set. Treat them as planning ranges, not prices.

Scenario 1: hold the line ($5 / $30). OpenAI keeps GPT-5.6 at GPT-5.5 rates and competes on capability alone. This is plausible if the upgrade is incremental and the pressure is competitive. Anthropic’s Claude Opus 4.8 sits at $5 / $25, close enough that holding the line keeps OpenAI price-comparable on the high end.

Scenario 2: double again ($10 / $60). If 5.6 delivers the rumored long-context and agentic gains, the 5.4-to-5.5 precedent says OpenAI will charge for them. A $10 / $60 standard rate would still undercut a GPT-5.5 Pro call, and it would mirror the way Anthropic priced Claude Fable 5 at $10 / $50, twice its own Opus tier. Frontier labs have been comfortable doubling.

Scenario 3: a tiered split. The most interesting outcome, and the one the leaks point toward. Reporting on a rumored “Codex UltraFast” mode suggests OpenAI may ship a cheaper, faster variant alongside the flagship, the way it has run mini and standard tiers before. That would mean two numbers: a low-latency rate aimed at high-volume coding loops, and a premium reasoning rate for the hard tasks. For cost-per-task math, a tiered split changes the calculation more than any single rate, because it lets you route cheap work to the cheap tier.

None of these is a forecast with a confidence interval. They are the three shapes the price can take, and each one has a real precedent from the last six months.

Why cost per task beats the sticker price

Here is the part most pricing pages miss. The per-token rate is the headline, but it is rarely the bill.

Picture two developers on the same model at the same $5 / $30 rate. One asks a question, gets an answer, moves on: a few thousand tokens, a fraction of a cent. The other runs an autonomous coding agent that reads twelve files, plans, edits, runs tests, reads the failures, and tries again, four times over. Same rate. The second task can cost a hundred times more, because agentic work multiplies tokens through loops, and output tokens (the expensive ones) stack up with every retry.

That is why a frontier price increase can matter less than it sounds. If GPT-5.6 doubles the rate but the rumored gains cut the number of retries a task needs, your bill can stay flat or fall. If it ships an UltraFast tier that runs the same loop at a fraction of the latency and rate, high-volume coding gets cheaper even as the headline number climbs. The method for turning token rates into a real per-task figure is the same one used in this breakdown of what Claude Code actually costs per task, and it applies to any model: estimate tokens per task, multiply by loop count, weight output heavily.

So when GPT-5.6 lands, the number to extract is not $X per million. It is dollars per task you actually run, on your actual workload.

Should you budget for GPT-5.6 now, or wait?

For a solo developer or a small team already on GPT-5.5, there is no reason to hold work for a model that may not arrive until the end of the month. GPT-5.5 at $5 / $30 is the live rate, and the cached-input discount is the lever that moves your bill today. The current rates across the major coding subscriptions sit on the AI pricing page if you are comparing plans rather than raw API tokens.

If you are planning a budget for the second half of 2026, plan against Scenario 2. Assume a doubling to $10 / $60 on the standard tier, then treat anything cheaper as upside. Budgeting against the worst plausible rate and being pleasantly surprised beats the reverse. If a tiered UltraFast option appears, revisit immediately, because routing the bulk of your token volume to a cheaper fast tier is the single biggest cost change on the table.

The one thing not worth doing is pricing your roadmap against a number nobody has published. Until OpenAI posts a rate card, every GPT-5.6 figure (including the three above) is a planning range.

When is GPT-5.6 coming out, and what is rumored?

OpenAI has confirmed nothing. The release-date expectation comes from coverage of the rumor cycle and from prediction markets such as Polymarket, where traders have priced a public launch by June 30, 2026 at roughly 80 to 89 percent.

On the feature side, the leak roundups point to deeper long-context reasoning (developer logs reportedly show context windows pushed toward 1.5 million tokens, up from GPT-5.5’s range), stronger multi-step planning and error recovery, and the Codex UltraFast latency mode. Every one of those is a rumor until OpenAI says otherwise. They matter here only because they are the features that would justify a price increase, which is why the doubling scenario is not a wild guess.

Frequently asked questions

How much does GPT-5.6 cost?

GPT-5.6 launched June 26, 2026 in three tiers per million tokens: Sol at $5 input / $30 output, Terra at $2.50 / $15, and Luna at $1 / $6. On July 30, 2026 OpenAI cut Terra 20 percent and Luna 80 percent and left Sol alone, so the current rates are Sol at $5 / $30, Terra at $2 / $12 and Luna at $0.20 / $1.20. Cached input bills 90 percent off ($0.50, $0.20 and $0.02 respectively). The flagship Sol holds GPT-5.5 pricing exactly.

Is GPT-5.6 more expensive than GPT-5.5?

No. The flagship, Sol, matches GPT-5.5 at $5 / $30, and the Terra and Luna tiers are cheaper. OpenAI held the flagship rate flat and competed on capability and a cheaper fast tier rather than raising the headline price.

What does GPT-5.5 cost per million tokens?

$5 input and $30 output on the standard tier, with cached input at $0.50. GPT-5.5 Pro is $30 input and $180 output. Long-context requests above 272K tokens move to $10 / $45.

Should I wait for GPT-5.6 or use GPT-5.5 now?

For most users, GPT-5.5 remains the practical choice for now: GPT-5.6 launched on June 26 in a government-gated limited preview open to only about 20 partners, with broader access planned in the following weeks. GPT-5.5 is shipping and priced, the cached-input rate cuts repeat-context cost sharply, and the cost-per-task math that decides your bill works the same on whichever model you run.

When was GPT-5.6 released?

OpenAI previewed GPT-5.6 (Sol, Terra and Luna) on June 26, 2026, but only to a limited set of about 20 government-approved partners via the API and Codex. OpenAI said wider public access is planned in the weeks after.

Sources

Get each breakdown before it makes the rounds

You get one email when a new source-backed analysis goes live: what AI agents actually cost, which models are worth running, and what the benchmarks really mean. No hype.

No spam. Unsubscribe anytime.

← Back to AI costs