Top Sites>Mmorphllm.com

User-Agent: *
Allow: /
Disallow: /admin
Disallow: /api

Host: https://www.morphllm.com
Sitemap: https://www.morphllm.com/sitemap.xml

# Morph

> Applied research lab that writes custom inference engines and custom GPU kernels to ship accelerated models purpose-built for code generation. Fast Apply, WarpGrep, Compact, and Glance — subagents and models that slot into any coding agent (Cursor, Claude Code, Cline, Windsurf) via OpenAI-compatible API or MCP.

The four products target the four places coding agents spend compute: applying edits, searching code, compacting context, and verifying changes. Speed numbers (10,500 tok/s Fast Apply, 33,000 tok/s Compact) come from the custom stack — not fine-tuning on a general-purpose serving layer. All APIs are OpenAI-compatible. Base URL: `https://api.morphllm.com/v1`.

## Products

- [Fast Apply](https://www.morphllm.com/products/fastapply.md): Merge AI code edits at 10,500 tok/s with 98% accuracy. Models `morph-v3-fast` (7B) and `morph-v3-large` (14B). 60x faster than full-file rewrites.
- [WarpGrep](https://www.morphllm.com/products/warpgrep.md): Code search subagent in a separate context window. 0.73 F1 in 3.8 steps, #1 on SWE-Bench Pro. No embeddings required. Model `morph-warp-grep-v2.1`.
- [Compact](https://www.morphllm.com/products/compact.md): Context compaction at 33,000 tok/s. Shrinks context 50-70% while keeping every surviving sentence verbatim. Enables 24+ hour agent sessions.
- [Glance](https://www.morphllm.com/products/glance.md): Browser agent that reads a PR diff, runs the affected UI flows, and posts video recordings back to GitHub.

## Reference

- [Home](https://www.morphllm.com/index.md): Overview of Morph and the four products.
- [Pricing](https://www.morphllm.com/pricing.md): Free tier (200 req/mo) plus usage-based pricing for every product.
- [Benchmarks](https://www.morphllm.com/benchmarks): SWE-Bench Pro, F1, and accuracy numbers.
- [Docs](https://docs.morphllm.com): SDK and API reference.
- [Agent context (llm-quickstart)](https://docs.morphllm.com/llm-quickstart): ~9k tokens of full Morph context, intended for coding agents to ingest.
- [Docs llms-full.txt](https://docs.morphllm.com/llms-full.txt): Complete Morph documentation concatenated as a single markdown file.
- [Quickstart](https://docs.morphllm.com/quickstart): First API call in under 2 minutes.
- [TypeScript SDK](https://docs.morphllm.com/sdk/quickstart): `@morphllm/morphsdk` — the main client for all products.
- [MCP setup](https://docs.morphllm.com/mcpquickstart): Model Context Protocol integration for Cursor, Claude Code, Windsurf, Cline, VS Code, Claude Desktop.
- [Authentication](https://docs.morphllm.com/auth): Bearer-token auth for the Morph API.
- [Self-hosting](https://docs.morphllm.com/api-reference/self-hosting): Run Morph models in your own environment.

## Writing on subagents and coding agent infrastructure

- [All Agents Will Be Coding Agents](https://www.morphllm.com/blog/all-agents-coding-agents.md): Why non-coding agent companies (Linear, Posthog, support, marketing, hardware telemetry) end up needing code execution.
- [WarpGrep v2](https://www.morphllm.com/blog/warpgrep-v2.md): SWE-Bench Pro results and why an isolated search subagent beats in-context search.
- [Compact SDK](https://www.morphllm.com/blog/compact-sdk.md): Compaction API deep-dive, why summarization fails, and patterns for 24+ hour agent sessions.
- [The Bitter Lesson for Coding Agents](https://www.morphllm.com/blog/bitter-lesson.md): Why specialized small models win on compute-bound sub-tasks.
- [Fast Apply and Fast Agents](https://www.morphllm.com/blog/fast-apply-fast-agents.md): Why apply latency is the bottleneck for agent feedback loops.
- [Best Practices for Building Coding Agents](https://www.morphllm.com/blog/best-practices.md): Patterns we've seen work across Morph customers.
- [The Code Search Bottleneck](https://www.morphllm.com/blog/code-search-bottleneck.md): Cognition's 60% search overhead measurement and what to do about it.
- [Multi-Agent Systems](https://www.morphllm.com/blog/multi-agent-systems.md): Anthropic's 90% multi-agent improvement and how we apply it at Morph.
- [Diffs vs Fast Apply](https://www.morphllm.com/blog/diffs-vs-fast-apply.md): Why lazy-edit merging outperforms unified-diff generation for agents.
- [Long Running Agents](https://www.morphllm.com/blog/long-running-agents.md): Context, compaction, and what breaks past hour 4.

## Pricing summary

Usage-based, no per-seat fees. Per 1M tokens:

- Fast Apply 7B: $0.80 input, $1.20 output
- Fast Apply 14B: $0.90 input, $1.90 output
- WarpGrep: $0.80 input, $0.80 output
- Compact: $0.20 input, $0.50 output
- Embeddings: $0.18
- Rerank: $0.10
- Router: $0.005 per request

Subscription plans (credits cover all APIs; 1 credit ≈ $0.00001):

| Plan | Price | Credits |
|------|-------|---------|
| Free | $0/mo | 250K |
| Starter | $20/mo | 2M |
| Pro | $60/mo | 8M |
| Scale | $400/mo | 80M |

## Optional

- [llms-full.txt](https://www.morphllm.com/llms-full.txt): Every page above inlined into a single stream.
- [Blog index](https://www.morphllm.com/blog): Full list of Morph engineering and research posts.
- [Contact](https://www.morphllm.com/contact): Sales, support, and enterprise inquiries.
- [Self-hosting](https://www.morphllm.com/blog/self-hosting.md): SOC2-compliant on-prem deployment.