Skip to content
FrontierHarness Eval

Open source alternatives to FrontierHarness Eval

200 open source apps like FrontierHarness Eval, closest match first.

Top picks

  1. C

    Claude Code Workflow

    Launch

    QuietHarness is a small set of working rules for AI coding agents such as Claude Code, Codex and Cursor. Its shared core asks the agent to check existing changes before it edits, keep unrelated changes intact, verify with real tests before

    FreeOpen source

  2. L

    Learn Claude Code

    Get

    Learn Claude Code is a teaching repository from shareAI-lab on harness engineering for real agents. Its central point is that an agent product equals a model plus a harness: the model supplies the agency, learned through training, and the h

    Open source

  3. S

    SDD Riper

    Launch

    Lightweight AI Agent Harness for agentic coding: let strong models explore while humans steer with minimal specs, checkpoints, approval, validation, and reverse sync.

    FreeOpen source

  4. A

    AgenTopology

    Launch

    Harness as code — the Terraform for AI agents. Define your agent team AND its memory once, deploy to Claude Code, OpenClaw, Cursor, Codex, Gemini, Copilot, Kiro. Declarative .at language + agent-maintained, Obsidian-portable company brains

    FreeOpen source

  5. A

    Agentacct

    Launch

    See what your coding agents did and what it cost. Breaks each task down into work steps — tools used, files changed, tests run, time and tokens spent. Local-first dashboard for Claude Code, Codex, OpenCode, and more. No login, no telemetry.

    FreeOpen source

  6. T

    ThinHarness

    Launch

    ThinHarness is a minimal, opinionated Python harness for purpose-built agents. It gives you the agent-loop primitives that are tedious to rebuild — scoped filesystem tools, structured output, retries, approvals, subagents, MCP, streaming, a

    FreeOpen source

More like FrontierHarness Eval

7
C

Claudexor

Multi-harness control plane for Claude Code, Codex, Cursor

Launch
8
A

ASSERT

Requirement-driven evaluation harness for AI agents and LLM applications

Launch
9
H

HarnessRouter

HarnessRouter Community Edition: the self-hosted

Launch
10
C

Cost Xray

See what Claude Code and Codex actually send to the API

Launch
11
E

Evalplus

Rigourous evaluation of LLM-synthesized code - NeurIPS 2023 & COLM 2024

Launch
12
H

Harnss

Open-source, desktop client/UI build to harness Claude Code

Launch
13
N

nanostack

A workflow harness that helps AI coding agents plan, review, test

Launch
14
P

Paseo

Self-hosted orchestrator for coding agents like Claude Code, Codex, and Cursor

Launch
15
L

LazyCodex

The one and only agent harness for complex codebases

Launch
16
O

Omnigent

Meta-harness that sits above Claude Code, Codex

Launch
17
O

ORG-2

System of record for how agents build software: Built-in rust harness & 20+ CLIs

Launch
18
K

Keesan12/martin-loop

Governed runtime for AI coding agents

Get
19
K

KARIMO

Claude Code harness engineering plugin — plan mode on steroids

Launch
20
A

Agtop

top-style TUI for monitoring AI coding agent sessions

Launch
21
H

Harness Score

Your AI coding agent is only as reliable as the harness around it

Launch
22
A

Archon

The first open-source harness builder for AI coding

Launch
23
H

Harness Books

Harness engineering in two books: constraints, query loops

Launch
24
P

Pi Harness

The operational superset of Pi Coding Agent

Launch
25
C

CodeBurn

Free, local tool to track AI coding token usage and cost across 37 tools

Launch
26
A

Agentglass

Every AI coding agent on your machine, on one screen

Launch
27
L

Litefuse

Agent Observability and Evaluation Platform

Launch
28
P

Paritok 4b V1

Non-destructive compression gateway for AI coding agents

Launch
29
B

Bromure

Proper sandboxing for agentic coding and web browsing

Launch
30
S

stagewise

The Agentic IDE for Open-Source Models

Launch
31
C

Codeman

Self-hosted mission control for AI coding agents: run Claude Code, OpenCode, Pi

Get
32
L

Little Coder

A harness optimized to smaller LLMs

Launch
33
H

HarnessKit

More than a skill manager — manage skills, MCP servers, plugins, hooks, CLIs…

Launch
34
C

CTX

CTX Fit finds the cheapest AI coding setup

Launch
35
C

cocode

Best ready-to-run DeepSeek Harness distribution: DSH desktop GUI

Launch
36
M

Meta_Kim

Governed execution layer for AI coding assistants: clarify intent

Launch
37
T

Trail-code

Turbocharge Claude Code, Cursor, Codex, Gemini & every coding agent: faster

Launch
38
C

Claudetop

htop for your Claude Code sessions — real-time cost, cache efficiency, model…

Launch
39
C

Codex

Lightweight coding agent that runs in your terminal

Launch
40
C

Codexmate

One dashboard for all your local AI coding agents

Launch
41
F

Forgecode

AI coding harness that runs inside your terminal, mixes LLMs per task

Launch
42
A

AGHub

One hub for every AI coding agent

Launch
43
B

BOSS Console

Open-source, multi-platform harness for AI agents

Launch
44
B

Benchmark Heaven

Model benchmarks & costs

Launch
45
A

Abtop

Like htop, but for AI coding agents

Launch
46
O

Opentraces

A local-first evidence layer for agent work, capturing what the agent sees

Launch
47
D

dev-3.0

Mission control for the One Person Studio

Launch
48
S

Superloopy

Lightweight Codex/Claude loop harness with strict evidence gates

Launch
49
B

Bettercode

Every coding agent in one desktop workspace

Launch
50
A

Agentsmith

Universal, model-agnostic operating harness for AI agents

Launch

About FrontierHarness Eval

F

FrontierHarness Eval

9 harness, same model, cost per pass varies 17x

FrontierHarness Eval is a public benchmark that compares AI coding agent harnesses while keeping the model fixed. It tested 9 harnesses in 12 configurations, among them Codex, Claude Code, Pi, Oh My Pi, Kimi Code, OpenCode, Hermes, Exo Harness and several DSH variants, on the same software engineering tasks. Every run used the Kimi K3 model on Runta, starting from an identical fresh checkpoint with the same CPU, memory and disk, for 360 trials in all. Results are shown for pass rate, median cost per task and per successful task, cache hit rate and time. The headline finding is that pass rates were close, roughly 50 to 67 percent, while cost per success differed by about 17 times. The evaluation code and a skill for testing your own harness are on GitHub.

Price
Free trial
Category
Developer Tools
Website
frontierharness.org