Kiro, built by AWS, is the official successor to Amazon Q Developer. Its standout feature isn't the number of models — it's how the Free tier gives you genuinely useful tools while drawing a sharp line around what you can't have: 50 credits per month, permanently free, no credit card required, with access to Claude Sonnet 4.5 and a solid set of open-weight models covering the full Spec/Vibe/Auto workflow. But GPT-5.6 and the entire Claude Opus line are locked behind the paywall.
This article is about the actual experience: what the Free tier lets you do, where it stops, and when you'll feel the upgrade urge. Prices below are list prices as of 2026-08-12. Check the Kiro website for the latest rates.
Free tier model breakdown: half in, half locked
Kiro Free covers roughly half of the 19-model lineup — Sonnet 4.5 and open-weight models are free, while GPT-5.6 and the entire Opus series sit behind paid plans. Kiro supports 19 models total (source). The Free tier covers roughly half:
| Category | Models | Free | Multiplier |
|---|---|---|---|
| Flagship | GPT-5.6 Sol / Terra / Luna | ❌ | 2.4× / 1.0× / 0.1× |
| Flagship | Claude Opus 5 / 4.8 / 4.7 / 4.6 / 4.5 | ❌ | 2.2× |
| Workhorse | Claude Sonnet 5 / 4.6 | ❌ | 1.3× |
| Workhorse | Claude Sonnet 4.5 / 4.0 | ✅ | 1.3× |
| Lightweight | Claude Haiku 4.5 | ❌ | 0.4× |
| Auto-router | Auto | ✅ | 1.0× |
| Open-weight | DeepSeek 3.2 | ✅ | 0.25× |
| Open-weight | MiniMax M2.5 / M2.1 | ✅ | 0.25× / 0.15× |
| Open-weight | GLM-5 | ✅ | 0.5× |
| Open-weight | Qwen3 Coder Next | ✅ | 0.05× |
This is a carefully designed split. What you get is more than enough to write working code — Sonnet 4.5 is still a very strong coder, DeepSeek 3.2 excels at long-context reasoning, and Qwen3 Coder Next outperforms many larger models on code completion.
What's locked are exactly the capabilities that create "overflow value": GPT-5.6 Sol's complex architecture design, Opus 5's 1M context for whole-codebase reasoning, Sonnet 5's agent-aware intelligence. These aren't about writing code faster — they're about tasks the Free tier simply can't attempt.
What 50 credits actually gets you, by model multiplier
Kiro credits deduct fractionally based on task complexity and model multiplier — the same credit pool can buy 50 tasks or 1,000, depending entirely on which model you pick. Kiro credits aren't "one per request." They deduct fractionally based on task complexity, weighted by model multiplier (source). Auto routing is the 1.0× baseline.
Your model choice directly determines how far 50 credits go. Estimated at 0.2–1.0 credits per typical coding task:
- Qwen3 Coder Next (0.05×): roughly 500–1,000 tasks. Insane value for completions, simple refactors, and comment generation. If 80% of your coding time is this kind of work, Free might genuinely be enough.
- Auto (1.0×): roughly 50 tasks. Kiro picks the model for you — convenient, but without paid access it can't reach GPT-5.6 or Opus, so you're mostly getting Sonnet-tier results.
- Sonnet 4.5 (1.3×): roughly 38 medium tasks. This is the Free tier workhorse — high code quality, solid reasoning. A couple uses per day and you'll last the month.
The real-world answer: if you're just opening Kiro for the odd file edit or a quick code question, 50 credits is plenty. But the moment you start running full Spec → generate → iterate cycles, you'll burn through them in days.
Spec, Vibe, and Auto are all available on Free — but the experience gap is real
Kiro's core differentiator — the three-workflow system (Spec, Vibe, Auto) — is fully open on Free, but the 50-credit cap means you can only sample, not integrate them into daily work. Kiro's core differentiation is its three-workflow system, all accessible on the Free tier (source).
Spec (spec-driven development): Write a requirement spec, and Kiro generates code while continuously validating against it. It's like having a CI/CD pipeline QA-ing your output. But a single complete Spec flow can eat 2–5 credits — on Free, you're not running many of these per month.
Vibe (conversational coding): Think Copilot-style back-and-forth. Good for exploration and small edits. Credit burn is manageable.
Auto (fully autonomous agent): Kiro reads your codebase, makes changes, runs tests — no step-by-step hand-holding. The best experience, the fastest credit drain — a long agent session can burn 5–10 credits.
The Free tier's awkward reality: all three modes are open, but the credit cap means you can only sample them, not integrate them into daily work. Pro's 1,000 credits is where these workflows stop being demos and start being tools.
Where the Free tier hits a wall: three moments you'll want to upgrade
The Free tier's real limit isn't model quality—it's three practical constraints: no GPT-5.6 or Opus access, no overage when credits run dry, and the psychological cost of constantly budgeting your AI usage. These are the three things that will push you toward upgrading:
No GPT-5.6 or Opus access. If you need large-scale architecture migration, extended context reasoning, or complex agent orchestration, Free's model selection won't cut it. This isn't a knock on Sonnet — it's that some tasks need Opus's 1M context window to model the full problem.
No overage option. When 50 credits run out, you're done. No $0.04/credit top-up, no partial extension. If you run dry on a Friday afternoon, you're either waiting until the monthly reset or upgrading immediately. The workflow interruption hurts more than the credit limit itself.
The psychological tax. This might be Free's biggest hidden cost — you start calculating every action: "Should I save credits and run this Spec later?" "Maybe use Qwen first, then switch to Sonnet for the final pass." When you're budgeting your AI usage, the efficiency gains start to slip.
When to upgrade? Not when you hit the credit limit — when you notice yourself compromising on work quality because of it. Once that happens, grab the $20 first-upgrade credit and go Pro.
Should you take the $20 credit and upgrade?
Take the $20 credit and try Pro—it's a free month with zero downside, and canceling costs nothing. AWS gives first-time paid upgraders a $20 subscription credit (via AWS Builder ID or social login) — essentially a free month of Pro.
The right way to use Free: spend a week running actual projects through Spec and Auto workflows. That's more reliable than any review. If by the end of that week the feeling of "not enough" outweighs "this is great," the tool is worth paying for — and the upgrade path is right there.
If a week goes by and you've barely dented the 50 credits, Free might genuinely cover your needs. But that probably also means you haven't pushed into Kiro's heavy-lift territory yet.