Augment CodevsCodeGPT
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
Augment Code
Augment Code · IDE extension
#32MCP
Panel
6.36 spec wins
- Reliability
- 6.3
- Usefulness
- 7.0
- Cost
- 5.2
- Longevity
- 6.5
“Charges the model's list price plus a 40 percent service fee, which is at least a markup you can read without a lawyer.”
CodeGPT
CodeGPT · IDE extension
#45MCP
Panel
6.35 spec wins
- Reliability
- 5.5
- Usefulness
- 6.5
- Cost
- 7.2
- Longevity
- 5.8
“The free plan gives you ten daily interactions on economy models, or unlimited if you bring the bill yourself.”
Spec by spec
| Spec | Augment Code | CodeGPT |
|---|---|---|
| Architecture | ||
| Category | IDE extension | IDE extension |
| Runs | local, cloud | local, cloud |
| Platforms | macos, linux, windows | macos, linux, windows |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | Yes | Yes |
| MCP server | Yes | No |
| Capabilities | ||
| Runs terminal commands | Yes | No |
| Multi-file edits | Yes | YesCodeGPT studies the codebase and proposes edits that you review before they are applied. |
| Git operations | Yes | No |
| Browser control | NoAuggie ships web-search and web-fetch tools only, so it can read pages but cannot drive a browser; browser automation requires adding a third-party MCP server. | No |
| Sandboxed execution | YesCosmos cloud environments are built from a Docker base image you pick from any public registry and each session starts from a fresh isolated snapshot; the local Auggie CLI has no sandbox, and self-hosted daemons isolate only via per-session Git worktrees. | No |
| Multi-agent orchestration | Yes | No |
| Headless / CI mode | Yes`auggie --print` runs non-interactively for scripts and CI, and Augment publishes GitHub Actions wrappers for PR descriptions and reviews. | No |
| Models | ||
| Backbone | Claude, GPT, Gemini, GLM, Kimi | Anthropic Claude, OpenAI GPT, Google Gemini |
| Bring your own model | NoModel choice is limited to Augment's curated list via /model or --model; no custom base URL, LLM gateway, or Bedrock/Vertex/Azure deployment is documented. | Yes |
| Local models | NoNo local or OpenAI-compatible endpoint is documented; inference always runs through Augment. | YesOllama and LM Studio are listed on the free plan alongside 15+ BYOK providers. |
| Cost | ||
| Pricing model | mixed | mixed |
| Starts at | $20/mo | $0/mo |
| Free tier | No | Yes |
| Bring your own key | NoLLM usage is billed by Augment as provider cost plus a 40% service fee; supplying your own provider key or cloud account is not documented. | Yes |
| Openness | ||
| Open source | No | No |
| License | proprietary | proprietary |
| GitHub stars | 283 | n/a |
Which one would each critic pick
| Critic | Augment Code | CodeGPT | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 7.0 | 6.8 | Augment Code — Use Augment when the codebase is the problem; on a small repo you are paying a markup for an index you do not need, so pick Claude Code there. |
| El Crítico | 6.0 | 6.0 | no preference |
| El Profesor | 6.5 | 6.0 | Augment Code — Two benchmark claims, one an ensemble and one a controlled comparison on the vendor's own harness; neither should be read against the public leaderboards. |
| La Inversora | 7.0 | 6.3 | Augment Code — Publishing a 40 percent service fee is either transparency or a confession, and exposing the index over MCP is a hedge toward becoming a component in other people's agents. |
| La Jefa | 6.8 | 5.8 | Augment Code — SOC 2 Type II, CMEK, audit trails and training exclusion on every plan, pooled usage to 50 seats, SSO and SCIM at Enterprise; approved with a spend ceiling. |
| El Hacker | 4.3 | 6.8 | CodeGPT — Fifteen-plus providers, a local runtime and a local model server both listed, and a tool-protocol client, which is most of what I want from a closed extension. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.