Current access
Checked October 3, 2026: Amp’s September 13 Free Agent announcement describes its Hobby tier as free when you bring your own model subscription or key and your own compute. Amp says there is no required monthly plan to use a ChatGPT subscription with Amp, and Hobby has no Amp token fees or limits. Amp also offers paid model inference without markup.
| Resource | Current route |
|---|---|
| Model access | Supply a supported subscription or key. The provider’s own billing, eligibility, and limits still apply. Amp also offers paid inference without markup. |
| Agent compute | Run agents on your own computer with Amp runners at no charge, or use Amp-hosted orbs on a pay-as-you-go basis. |
| Amp plan | Hobby is listed as Free; remote orbs, Amp model inference, and a paid model subscription or API key can still cost money. |
The announcement does not describe a repeatable $10/day model grant. The old grant math, model roster, and account-specific entitlement below are historical. A particular account’s subscription compatibility, model access, quota, successful inference, and agent outcomes were not tested.
Historical snapshot: February 2026 grant, models, and setup
The calculations below are retained as dated product history, not as a current free-model recommendation. For current model selection, use Opus 5.5, GPT-6 Sol and Luna, GLM-5.2, and Kimi K2.7 Code.
Why This Tool
Most free AI coding tools have catch-22s: credits expire, previews end, or you get a lesser model. AMP’s ad-supported tier is different: predictable daily Opus 4.5 access that renews every 24 hours, no credit card required, no expiration anxiety.
The sustainable advantage:
- $10/day grant: Covers ~2M input tokens or ~400K output tokens of Opus 4.5
- Daily refresh: Quota resets every 24 hours—stackable if you plan ahead
- Non-intrusive ads: Banner ads only from DevTools companies (Axiom, Chainguard, Vanta)—no video pre-rolls
- No credit card: Unlike Kiro, you don’t need to hand over payment info
- No preview uncertainty: Unlike Antigravity, this isn’t a “terms may change” situation
The trade-off: Ads in your development environment. For some, this is unacceptable. For others, it’s a fair price for $200/month model access.
What You Get
Free Tier: Ad-Supported
Daily Grant: $10/day (resets every 24 hours)
| Model | Input Tokens | Output Tokens | Cost per 1M |
|---|---|---|---|
| Claude Opus 4.5 | ~2,000,000 | ~400,000 | $5 in / $25 out |
| GPT-5.2 | ~5,700,000 | ~2,000,000 | $1.75 in / $5 out |
| Claude Sonnet 4 | ~3,300,000 | ~1,400,000 | $3 in / $7 out |
| Fast models | ~10,000,000+ | ~5,000,000+ | $0.50-$1/1M |
What this means in practice:
- ~50-100 complex coding requests with Opus 4.5 per day (depending on output length)
- ~200-300 smaller tasks with Sonnet or GPT-5.2
- All agentic features included — Full edit-test-loop capabilities
- 200K context window — Full codebase awareness
- Thread sharing — Collaborate with team members
Best for: Developers who want sustainable, ongoing Opus 4.5 access without managing credits or worrying about preview terms changing. Heavy enough for daily professional use if you manage the quota.
Alternative: $10/Day Grant (Same Tier)
The free tier IS the $10/day grant—there’s no separate signup. The grant mechanics:
- Automatic model selection — AMP picks the right model for each task
- Smart mode — Uses Opus 4.5 for complex tasks, cheaper models for simple edits
- Rush mode — Faster, cheaper—good for straightforward edits
Best for: Users who trust AMP’s routing vs. manually selecting models.
The Ad-Supported Model: Innovation or Intrusion?
AMP’s ad-supported approach is unique in frontier AI coding tools—and controversial.
How the Ad Model Works
What you see:
- Banner ads at the bottom of the editor and CLI interface
- Advertisers are DevTools companies: Axiom, Chainguard, Vanta, WorkOS
- No code snippets shared with ad partners (confirmed by CEO Quinn Slack)
The math that makes it work:
- Direct Opus 4.5 API cost: ~$15-30/day for typical usage
- AMP’s ad-supported cost to you: $0 (ads subsidize the difference)
- Your attention value: ~$10-20/day to advertisers in the DevTools space
The Controversy
Privacy concerns (partially resolved):
| Issue | Original Policy | Current Policy (Jan 2026) |
|---|---|---|
| Training data opt-in | Required for free tier | Removed — no longer required |
| Thread storage | Sourcegraph servers | Still required (for collaboration) |
| Enterprise use | Allowed | Disabled — compliance restriction |
Developer sentiment:
- Pro: “Kind of fascinating… could quickly become appealing once we collectively realize the untenable costs of true agentic coding” — Kaushik Gopal, Instacart Principal Engineer
- Con: “Ads in my IDE feels wrong. I pay to remove ads everywhere else—why would I invite them into my workflow?”
The verdict: The ad model is polarizing but functional. Training data opt-in removal (Jan 2026) addressed the biggest privacy concern. Enterprise teams still need paid workspaces.
Why It Matters for the Industry
AMP is testing a hypothesis: Can ad-supported models sustainably fund frontier AI access?
- If successful: Expect copycats. Free tiers could shift from “loss leader” to “ad-supported” across the industry.
- If failed: Validates subscription models. Proves developers won’t tolerate ads in professional tools.
Either outcome shapes how you access AI coding tools in 2026-2027.
The Setup
- Visit ampcode.com
- Select “Free Tier (Ad-Supported)”
- Sign up for free account (no credit card)
- Install VS Code extension or use web interface
- Daily $10 quota applied automatically
- Open a project and describe what you want:
"Add user authentication with JWT tokens" - AMP implements it autonomously
Time to first agentic workflow: 5 minutes
The Catch
Daily Limits Are Real
$10/day sounds generous until you hit it:
- Heavy Opus 4.5 sessions (long outputs) burn through quota fast
- Complex refactoring can consume 50-100K output tokens = $1.25-2.50
- 10-20 complex requests per day is realistic, not unlimited
Quota management required:
- Check remaining balance with
amp status - Switch to “rush” mode for cheaper tasks
- Plan intensive work across multiple days
Ads Are Present (Obviously)
Non-intrusive but there:
- Banner ads don’t interrupt workflow
- Still: you’re looking at ads while coding
- Alternative if you can’t tolerate ads: Antigravity (free, no ads, preview status)
Enterprise Restrictions
Free tier disabled for enterprise workspaces:
- Compliance teams can’t use ad-supported model
- Thread storage on Sourcegraph servers may violate policies
- Enterprise path: Paid workspaces only
Model Availability Caveats
Daily allocation means pacing:
- Morning Opus 4.5 binge → afternoon throttling to cheaper models
- Quota shared across all models (not per-model limits)
- Heavy users will hit the cap consistently
Competitive Positioning: How AMP Stacks Up
Free Tier Comparison
| Tool | Free Access | Model | Credit Card | Ads | Sustainability |
|---|---|---|---|---|---|
| AMP | $10/day (~400K Opus out) | Opus 4.5 | No | Yes | Ongoing |
| Antigravity | Session-limited | Opus 4.5 | No | No | Preview (uncertain) |
| Kiro | 500 credits/month | Sonnet 4.5 only | Yes | No | 30-day expiry |
| OpenCode Zen | Unlimited* | Qwen 3.6 Pro | No | No | Ongoing |
| Kilo Code | BYOK only | Your keys | No | No | Depends on your API |
Key differentiator: AMP is the only tool offering sustainable, ongoing Opus 4.5 access (the $200/month model) without subscription. The ad trade-off is explicit and functional.
vs Paid Alternatives
| Tool | Monthly Cost | Opus 4.5 | Key Advantage |
|---|---|---|---|
| AMP (daily max) | $0 | ~400K tokens/day | Free, sustainable |
| Claude Pro | $20/mo | Limited | Official Anthropic, no ads |
| Claude Max | $200/mo | Unlimited | Unlimited Opus 4.5, no ads |
| Cursor Pro | $20/mo | Via API | Full IDE experience |
| Cursor Business | $40/mo | Via API | Team features |
Break-even analysis:
- If you use <400K Opus 4.5 output tokens/month → AMP is cheaper than Claude Pro ($20)
- If you use >12M Opus 4.5 output tokens/month → Claude Max ($200) is cheaper
- Most developers: AMP covers 80-90% of needs for $0
When to Choose What
Choose AMP if:
- You want ongoing Opus 4.5 without subscription
- You can tolerate banner ads
- Your usage is sporadic (<$10/day equivalent)
- You value thread sharing/collaboration
Choose Antigravity if:
- You can’t tolerate any ads
- You accept preview status uncertainty
- You want Google’s ecosystem
Choose Claude Pro if:
- You need enterprise compliance
- Ads are a non-starter
- You want official Anthropic support
Choose Claude Max if:
- You consistently exceed $10/day usage
- Unlimited Opus 4.5 is essential
- Cost is secondary to workflow
Where Next
Hit your daily $10 limit? → Switch to rush mode for cheaper processing → Antigravity — Free Opus 4.5 with different limits (no ads) → NVIDIA NIM — Kimi K2.6 internal evaluation trial; not a production fallback → Upgrade to Claude Pro ($20/mo) for predictable unlimited
Want no ads at all? → Antigravity — Free Opus 4.5, no ads (public preview) → OpenCode Zen — Limited-time free models; verify the live roster and data terms → Claude Pro ($20/mo) — Official, no ads
Need more control over implementation? → OpenCode Zen — Direct model access, you steer every change → Claude Code — Terminal-native, direct Anthropic access
Want VS Code native integration? → Cursor — $20/mo, full IDE experience, no ads → Kilo Code — BYOK mode for flexibility
Enterprise/compliance requirements? → Claude Pro/Max — Enterprise-grade terms → Self-hosted options — Full control, BYOK
Best Use Cases
Perfect for:
- Refactoring mature codebases (Opus 4.5 excels here)
- Adding features to existing projects
- Test-driven development (AMP writes tests + implementation)
- Learning from AI implementation patterns
- Daily coding with predictable Opus 4.5 access
- Teams who want thread sharing without setup
Not ideal for:
- Greenfield projects (too many decisions to automate)
- Highly specific implementation requirements
- Real-time pair programming feel
- Users who can’t tolerate any ads
- Enterprise environments (compliance restrictions)
- Consistently heavy usage (>400K output tokens/day)
The Free Tier Landscape: February 2026
Your options for Opus 4.5-level reasoning:
| Option | Cost | Model | Limitation |
|---|---|---|---|
| AMP | $0 | Opus 4.5 | Ads, $10/day |
| Antigravity | $0 | Opus 4.5 | Preview status |
| Kiro | $0 | Sonnet 4.5 only | 500 credits, 30 days |
| OpenCode Zen | $0 | Qwen 3.6 Pro | Strong coding, 1M context |
The reality: You can get frontier-level AI coding without paying—but each option has trade-offs. AMP’s trade-off is ads for sustainability. That’s a viable choice for many developers.
Related links
Value & Pricing:
- Free Frontier Stack — Complete free tier guide with all options
- Smart Spend Guide — When to upgrade from free to paid
- Compare Models — Budget, mid-range, premium tier breakdowns
Alternative Tools:
- Antigravity — Free Opus 4.5 without ads (preview)
- NVIDIA NIM — Kimi K2.6 internal evaluation trial; production requires a separate subscription
- OpenCode Zen — Limited-time free models in an open-source coding harness
- Kiro — 500 credits for Sonnet 4.5 (no Opus on free)
- Kilo Code — BYOK mode
Model Deep Dives:
- Claude Opus 4.5 — Capabilities and benchmarks
- Kimi K2.5 — Historical model and access evidence
Industry Analysis:
- Kimi k2.5 Pricing Strategy — Why free tiers exist and where they’re heading
- Codex vs Claude vs Kimi Comparison — Agent tool landscape
Archived-source check: February 3, 2026. These sources support the historical snapshot above, not current access.
Sources: ampcode.com/manual, ampcode.com/chronicle, tessl.io/blog/amp-ad-model