Key takeaways
- Amp spun out of Sourcegraph as an independent company, Amp, Inc., in December 2025 — Quinn Slack and Beyang Liu lead it as a frontier agent lab
- Multi-model by design: Opus 4.8 for smart mode, GPT-5.5 for deep/rush, Gemini for review and search — Amp picks the model, you don't
- The ad-supported Amp Free experiment hit a $10M+ annual run rate, then Amp killed the ads in March 2026 and kept the $10/day allowance
- Opinionated design philosophy: features get killed if the team doesn't love them — public thread sharing and the editor sidebar are recent casualties
FAQ
What is Amp?
Amp is a frontier coding agent for terminal and editor, built by Amp, Inc. (spun out of Sourcegraph in December 2025), using multiple AI models optimized for different coding tasks.
How much does Amp cost?
Free tier with a $10/day usage allowance (no longer ad-supported as of March 2026, though allowances are being reduced for less-active users). Then pay-as-you-go with zero markup on provider API pricing for individuals; $5 minimum credit purchase.
What models does Amp use?
As of June 2026, Claude Opus 4.8 powers smart mode, GPT-5.5 powers deep and rush modes plus the Oracle, and Gemini models handle review and search. Amp selects models for you — there is no user model picker, and BYOK is enterprise-only.
Who competes with Amp?
Claude Code, Cursor, Factory Droids, and Tembo for orchestration.
Executive Summary
Amp began as Sourcegraph's entry into the AI coding agent market and spun out as an independent company, Amp, Inc., in December 2025 — Sourcegraph co-founders Quinn Slack and Beyang Liu now run Amp as a "frontier research lab" while Dan Adler leads the code-search business. [1] Amp positions itself as a "frontier coding agent" with a multi-model approach and an aggressively opinionated design philosophy. The free tier ($10/day allowance) and transparent pay-as-you-go pricing differentiate it from subscription-based competitors — and its short-lived ad-supported model reached a $10M+ annual run rate before Amp shut the ads off in March 2026. [2]
| Attribute | Value |
|---|---|
| Company | Amp, Inc. (spun out of Sourcegraph, Dec 2025) [1] |
| Founded | Amp launched 2025; Sourcegraph founded 2013 |
| Funding | Sourcegraph raised $223M; Amp, Inc. has not disclosed standalone funding — reportedly already profitable [3] |
| Leadership | Quinn Slack (CEO), Beyang Liu (co-founder) |
| Headquarters | San Francisco, CA |
Product Overview
Amp is a terminal-first coding agent with editor extensions, a web interface, and mobile apps. [4]
The core philosophy is opinionated design: "If we don't use and love a feature, we kill it." Recent casualties include Tab completion, public thread sharing (removed June 2026), and — per the team's own podcast — the editor sidebar extension, which they announced "will soon self-destruct." [5][6]
In May 2026 the next-generation "Amp Neo" client became simply Amp for all users, the npm package moved to @ampcode/cli, and the team announced Amp Labs, a program connecting top developers with enterprises. [5]
Key Capabilities
| Capability | Description |
|---|---|
| Multi-Model Routing | Amp picks the best model per task — no user model picker |
| Sub-agents | Oracle (deep reasoning), Librarian (codebase Q&A), Review, Painter |
| Threads | Version-controlled conversation history, workspace-level sharing |
| AGENTS.md Support | Project-specific guidance files |
| Three Modes | Smart, Rush, and Deep for different workloads |
| Plugins | Plugin system with UI on web, synced with the TUI |
Product Surfaces / Editions
| Surface | Description | Availability |
|---|---|---|
| Terminal CLI | Primary interface (@ampcode/cli), Mac/Linux/WSL/Windows | GA |
| VS Code Extension | Editor integration (team has signaled the sidebar's days are numbered) | GA |
| JetBrains Extension | IDEs 2025.1+ | GA |
| Neovim / Zed | For terminal purists | GA |
| Web Interface | Thread sharing, remote control, plugin UI | GA |
| Mobile Apps | Agents accessible from mobile | GA |
Technical Architecture
Amp operates in three modes: [7]
- Smart mode — Unconstrained state-of-the-art model use (Claude Opus 4.8 as of June 2026)
- Rush mode — GPT-5.5 with no reasoning, for small well-defined tasks ("twice as fast")
- Deep mode — Extended reasoning with GPT-5.5 for complex problems
Key Technical Details
| Aspect | Detail |
|---|---|
| Deployment | Cloud-based with local CLI |
| Model(s) | Claude Opus 4.8 (smart), GPT-5.5 (deep/rush/Oracle), Gemini 3.1 Pro (review), Gemini 3 Flash (search), Claude Sonnet 4.6 (Librarian), GPT Image 2 (Painter) [8] |
| Model choice | Amp-selected only; BYOK limited to enterprise regional endpoints [7] |
| Integrations | VS Code, JetBrains, Neovim, Zed, MCP |
| Open Source | No (proprietary) |
Category fit note: Amp is model-agnostic at the vendor level — it routes across Anthropic, OpenAI, and Google models — but the user never chooses the model and there is no general BYOK. It qualifies as multi-provider, with the caveat that the agnosticism is Amp's, not yours.
Strengths
- Frontier-lab focus — Independent company shipping on fast cycles, no longer tied to Sourcegraph's enterprise cadence
- Multi-model approach — Routes across Anthropic, OpenAI, and Google models per task, automatically
- Transparent pricing — Zero markup on provider API pricing for individuals
- Opinionated quality — Willing to remove features (and even revenue lines like ads) that don't fit
- Reportedly profitable — Rare among coding-agent startups [3]
- Generous free tier — $10/day allowance enables real evaluation, now without ads
Cautions
- Expensive in practice — Pass-through pricing on frontier models adds up; cost is the most common complaint in community threads [9]
- No model control — You can't pick models or bring your own key (BYOK is enterprise-only); you trust Amp's routing
- Breaking changes — "No backcompat, no legacy features" is a double-edged sword; public thread sharing and the editor sidebar were both axed in 2026
- Free tier in flux — Ads are gone, but daily allowances are being paused or reduced for less-active users and older clients [2]
- Newly independent — Amp, Inc. is months old as a standalone company with no disclosed funding of its own
What Developers Say
From the Hacker News discussion of Amp's spinout (December 2025): [9]
"Amp is still on top, costs more, but has the best results. The librarian and oracle make it leagues ahead of the competition." — incoming1211
"Even if I find it great, it's way too expensive, and having a number tick up/down makes me nervous." — embedding-shape
"They don't have their own model so what are you really paying for?" — lvl155
"Amp blew me away and was my primary workhorse. But then I switched to Claude CLI and that was good enough and significantly cheaper/faster." — sergiotapia
The pattern as of June 2026: strong praise for output quality and the Oracle/Librarian sub-agents, persistent friction over usage-based cost versus flat-rate subscriptions.
Pricing & Licensing
| Tier | Price | Includes |
|---|---|---|
| Free | $0 | $10/day allowance; ad-free since March 2026; interactive use only; allowances being reduced for less-active users [2] |
| Pay-as-you-go | Variable | Zero markup on provider API pricing; $5 minimum credit purchase; credits expire after one year of inactivity [7] |
| Enterprise | Custom | SSO/SAML, zero data retention, BYOK regional endpoints, MCP allowlists, analytics API |
Licensing model: Freemium with usage-based pricing
Hidden costs: None on markup — but frontier-model usage at cost can exceed flat-rate competitor subscriptions
Competitive Positioning
Direct Competitors
| Competitor | Differentiation |
|---|---|
| Claude Code | Claude Code is Anthropic-only; Amp routes across providers |
| Cursor | Cursor is IDE-native; Amp is terminal-first with lighter editor support |
| Factory AI | Factory has enterprise Droids; Amp is developer-focused |
| Codex | Codex is OpenAI-only; Amp uses multiple providers |
When to Choose Amp Over Alternatives
- Choose Amp when: You want curated multi-model capability with transparent pricing and a terminal-first workflow
- Choose Claude Code when: You're committed to Claude models and want official Anthropic support (or flat-rate subscription economics)
- Choose Cursor when: You prefer a full AI IDE over a terminal agent
- Choose Factory when: You need enterprise-grade Droids with workflow automation
Ideal Customer Profile
Best fit:
- Terminal-native developers who prefer CLI over GUI
- Engineers wanting latest models without subscription lock-in
- Developers who value opinionated, curated tool experiences — and are happy to delegate model choice
- Budget-conscious evaluators who can live within the $10/day free allowance
Poor fit:
- Developers who want to pick their own models or bring their own keys
- Users who want stable, backward-compatible tooling
- Heavy users for whom usage-based frontier pricing exceeds flat-rate subscriptions
- Organizations requiring single-vendor model relationships
Viability Assessment
| Factor | Assessment |
|---|---|
| Financial Health | Solid — reportedly profitable at spinout; willing to kill a $10M+ ad run rate on principle [2] |
| Market Position | Challenger — respected by power users, smaller mindshare than Claude Code/Cursor |
| Innovation Pace | Rapid — opinionated updates, monthly model swaps, fast feature churn |
| Community/Ecosystem | Growing — vocal HN following; Amp Labs program for enterprises |
| Long-term Outlook | Positive but watch the spinout — independence brings focus and standalone-funding risk |
Investors from Sourcegraph's board (a16z, Sequoia, Redpoint, Craft, Goldcrest) continue to serve on the boards of both companies. [1] The risk is whether the terminal-first, no-model-choice, usage-priced approach resonates broadly against flat-rate, pick-your-model competitors.
Bottom Line
Amp has graduated from Sourcegraph side-project to independent frontier-agent lab. The multi-model routing and zero-markup pricing are compelling, and the team's willingness to kill features — and even a $10M ad business — signals conviction. But you cede model choice entirely to Amp, usage-based costs run hot for heavy users, and the free tier is tightening.
Recommended for: Terminal-native developers who want curated multi-model capability with transparent, usage-based pricing and appreciate opinionated tool design.
Not recommended for: Teams wanting model choice or BYOK, stable backward compatibility, or predictable flat-rate costs.
Outlook: As an independent company, Amp will keep moving fast — Opus 4.8 landed in smart mode within days of release, and surfaces now span CLI, web, and mobile. [5][6] Expect continued feature churn and a free tier that gets more conditional. The curated multi-model approach positions it well as the model landscape fragments — for developers willing to let Amp drive.
Research by Ry Walker Research • methodology
Sources
- [1] Why Sourcegraph and Amp are becoming independent companies
- [2] Amp Free Is Ad-Free (Amp News, March 2026)
- [3] Sourcegraph spins out AI coding agent Amp as a standalone company
- [4] Amp Homepage
- [5] Amp Chronicle (News)
- [6] Raising an Agent Podcast
- [7] Amp Owner's Manual
- [8] Amp Models
- [9] Hacker News: Amp, Inc. — Amp is spinning out of Sourcegraph