OpenAI GPT-5.6 in Kiro: Pricing & 82% Cost Claim

OpenAI’s full GPT-5.6 family is now available in Kiro, AWS’s software-development agent. Paid Kiro users can select Sol, Terra, or Luna for planning, implementation, review, testing, and other codebase-aware work.

The launch changes model choice inside Kiro, not OpenAI’s API rate card. Kiro meters work in product credits, while direct API users pay by tokens. The live table and chart above pull current direct-provider rates from our dataset so the two purchasing models are not confused.

What changed in Kiro

This is the first OpenAI model family offered in Kiro. All three GPT-5.6 tiers support Kiro’s spec-driven workflow: turning requirements into technical designs and tasks, executing multi-step changes, reviewing checkpoints, and checking implementations with property-based tests.

Kiro’s model documentation says GPT-5.6 is available on its paid tiers, not the Free tier. It lists the models for US and EU customers, but says GPT-5.6 inference is served from the US even when a user’s Kiro profile is in Europe. Teams with data-residency requirements should review that routing before enabling the models.

Kiro’s paid plans currently bundle the same effective $0.02 per included credit at different capacity levels. Add-on credits cost $0.04 each, so teams that repeatedly exceed the plan allowance pay twice the bundled unit price.

Kiro plan	Monthly price	Included credits	Add-on price
Pro	$20 per user	1,000	$0.04 per credit
Pro+	$40 per user	2,000	$0.04 per credit
Pro Max	$100 per user	5,000	$0.04 per credit
Power	$200 per user	10,000	$0.04 per credit

The Free tier includes 50 credits but does not list GPT-5.6 access. Plan credits reset monthly; unused included credits do not roll over. Purchased add-on credits roll over and expire 12 months after purchase.

GPT-5.6 credit comparison in Kiro

Kiro model	Credit multiplier	Context shown by Kiro	Best starting use
GPT-5.6 Luna	0.1x	272K	Repeated, high-volume agent work
GPT-5.6 Terra	1.0x	272K	Routine multi-step development
GPT-5.6 Sol	2.4x	272K	Hard refactors and long-horizon terminal work

On the published multipliers, Terra draws ten times Luna’s base credit factor and Sol draws 24 times Luna’s. That does not guarantee an identical prompt always has those exact ratios. Kiro says actual credit use also depends on generated tokens, internal thinking depth, tokenizer behavior, and the duration of the task.

Kiro’s Auto option remains the sensible default when a team has not yet built model-specific evaluations. Manual routing becomes useful after accepted-task data shows that a higher multiplier prevents enough retries or corrections to earn back its extra credits.

What the 82% cost claim means

OpenAI says joint testing with AWS found that GPT-5.6 Terra completed successful tasks on Terminal-Bench 2.1 in Kiro at roughly 82% lower cost. It attributes the result to GPT-5.6’s price-performance and Kiro’s structured requirements, design, and task context reducing missteps.

The announcement does not name the comparison model, starting configuration, sample size, pass rate, token totals, or whether the figure uses Kiro credits or an internal inference cost. Treat 82% as a vendor-reported benchmark result—not a forecast for a repository or invoice.

The right production metric is cost per accepted task. A cheaper run that needs more retries, review, or repair can lose to a higher-multiplier model that finishes correctly once.

Who benefits—and who may pay more

Teams already using Kiro gain the most: they can evaluate GPT-5.6 without building their own agent harness, context pipeline, or spec workflow. Luna is the clearest candidate for repeated maintenance and lightweight tasks; Terra targets balanced daily development; Sol is for the hardest work where extra planning and tool coordination matter.

Sol-by-default teams may burn through Kiro credits faster without a matching quality gain. European organizations may also rule out the family when US inference conflicts with policy. Direct API builders get no new discount from this integration and still need to fund orchestration, tools, storage, and execution around the token bill.

Compare the model card on our OpenAI pricing page with current Anthropic pricing, then test your own input-output mix in the AI token calculator. Our earlier GPT-5.6 builder guide covers routing, caching, compaction, and lower reasoning effort outside Kiro.

Teams comparing a separate fixed-quota coding option can also benchmark the Z.ai coding plan on the same repository task and acceptance tests.

Affiliate disclosure: AI Pricing Guru may earn a commission from the sponsored Z.ai link at no extra cost to you. It does not affect this analysis.

What Kiro users should do now

1. Run the same bounded repository task with Auto, Luna, Terra, and Sol.
2. Record credits used, wall-clock time, tests passed, review changes, and retries for each run.
3. Start routine work on Luna or Auto; move to Terra when consistency or multi-step depth improves the accepted-task cost.
4. Reserve Sol for failures, complex refactors, and terminal work where it produces a measurable quality gain.
5. Confirm US inference is compatible with the project’s data and compliance requirements.
6. Compare Kiro’s credit result with the direct API calculator above only as two different buying paths; Kiro includes product orchestration that a token estimate does not.

Bottom line

GPT-5.6 in Kiro gives developers a useful three-step cost ladder inside a structured coding agent. Luna’s 0.1x multiplier makes it the efficiency choice, Terra is the balanced tier, and Sol should be an escalation path rather than an automatic default.

The reported 82% reduction is promising but underspecified. Validate it on accepted changes in your own codebase before revising budgets or routing policy.

Sources: OpenAI’s GPT-5.6 in Kiro announcement, Kiro’s model and credit documentation, Kiro’s pricing page, and the live AI Pricing Guru dataset. Availability, multipliers, routing, and pricing details verified August 24, 2026.