# Feedback Bench 1.1: Coding Agents, Ranked by their users’ feedback

Built 2026-10-19. Window 2026-08-31 to 2026-10-18. Web page: https://feedbackbench.com/coding-agents/. Built by Enterpret (https://www.enterpret.com), which runs the same feedback analysis on a company's own customer feedback.

Feedback Bench 1.1 ranks 17 coding agents from what their users say in public: Reddit, X, G2 and Trustpilot. Every post is labelled against 63 criteria in 9 areas. In Feedback Bench 1.1, popularity and customer love count equally: the Feedback Score is 50 + 10 × z(½ z(Popularity) + ½ z(Customer love)), where each z-score is the distance from the average agent in standard deviations. 50 is an agent that is average on both. Popularity is how many people post about the agent (log share of users, from the least-discussed agent at 0 to the most-discussed at 1); Customer love is how those posts rate it against the category (0.5 = category norm). Every week since 2026-08-31 counts, and recent weeks count more: a week's weight halves every 4 weeks. Full method: [https://feedbackbench.com/coding-agents/method.md](https://feedbackbench.com/coding-agents/method.md).

Also ranked: [personal agents](https://feedbackbench.com/index.md).

## Why this benchmark

Lab benchmarks test an agent on a fixed set of tasks. Feedback Bench ranks 17 coding agents on what their users say about them in public.

- **What it measures.** Large language models label each post against 63 criteria in 9 areas. Each agent gets a score on every criterion. The posts come from Reddit, X, G2 and Trustpilot.
- **How it counts.** The unit is the user-week: one person, about one agent, in one week. Twenty posts by one person in one week count once. The Feedback Score gives equal weight to two parts: how many people post about the agent, and how they rate it against the category norm. Agents with few posts are pulled toward that norm.
- **Sources and data.** Every number links to the posts behind it, quoted word for word, with a link to the original. The method and the data are public as Markdown, JSON and CSV.
- **Coding agents.** Coding agents are the first category. The ranking updates every week. Personal agents are the second.

## Ranking

Rank is the position by Feedback Score; the rank range column gives the ranks each agent reaches in 95% of bootstrap resamples. Criteria ranking high / low counts, of the 63 criteria, those where the agent clearly sits in the top / bottom half. Position on a criterion: its place among the agents with at least 30 rated user-weeks there, ordered by customer love. It ranks high when both its position and its 95% interval put it in the top half, low when both put it in the bottom half; neither otherwise, or when fewer than 4 agents qualify.

| Rank | Agent | Maker | Feedback Score | 95% interval | Rank range | Popularity | Customer love | Customer love 95% interval | Criteria ranking high / low |
|---|---|---|---|---|---|---|---|---|---|
| 1 | [Claude Code](https://feedbackbench.com/coding-agents/agents/claude-code.md) | Anthropic | 68.5 | 67.4–69.8 | 1–1 | 1.000 | 0.573 | 0.564–0.582 | 30 / 9 |
| 2 | [Devin](https://feedbackbench.com/coding-agents/agents/devin.md) | Cognition | 64.2 | 62.4–65.8 | 2–4 | 0.638 | 0.611 | 0.595–0.626 | 12 / 2 |
| 3 | [OpenCode](https://feedbackbench.com/coding-agents/agents/opencode.md) | Anomaly (open source) | 63.2 | 61.7–64.5 | 2–4 | 0.844 | 0.562 | 0.550–0.571 | 20 / 8 |
| 4 | [Pi](https://feedbackbench.com/coding-agents/agents/pi.md) | Earendil Works (open source) | 62.4 | 60.7–63.8 | 2–4 | 0.641 | 0.596 | 0.582–0.610 | 16 / 1 |
| 5 | [Amp](https://feedbackbench.com/coding-agents/agents/amp.md) | Amp | 56.0 | 54.2–57.8 | 5–6 | 0.482 | 0.578 | 0.563–0.593 | 11 / 1 |
| 6 | [Factory](https://feedbackbench.com/coding-agents/agents/factory.md) | Factory | 55.1 | 52.8–57.2 | 5–7 | 0.475 | 0.572 | 0.553–0.591 | 9 / 0 |
| 7 | [Cursor](https://feedbackbench.com/coding-agents/agents/cursor.md) | Anysphere | 52.9 | 51.4–54.4 | 6–7 | 0.802 | 0.487 | 0.474–0.500 | 6 / 20 |
| 8 | [Zed](https://feedbackbench.com/coding-agents/agents/zed.md) | Zed Industries | 48.7 | 47.4–50.1 | 8–10 | 0.541 | 0.507 | 0.496–0.519 | 2 / 4 |
| 9 | [GitHub Copilot](https://feedbackbench.com/coding-agents/agents/copilot.md) | GitHub | 48.1 | 46.5–49.9 | 8–11 | 0.557 | 0.499 | 0.485–0.514 | 2 / 7 |
| 10 | [OpenAI Codex](https://feedbackbench.com/coding-agents/agents/codex.md) | OpenAI | 47.7 | 46.6–48.8 | 8–11 | 0.980 | 0.409 | 0.400–0.417 | 10 / 27 |
| 11 | [Cline](https://feedbackbench.com/coding-agents/agents/cline.md) | Cline (open source) | 46.7 | 45.2–48.3 | 9–11 | 0.497 | 0.500 | 0.487–0.514 | 1 / 6 |
| 12 | [Conductor](https://feedbackbench.com/coding-agents/agents/conductor.md) | Melty Labs | 43.3 | 42.4–44.3 | 12–13 | 0.305 | 0.512 | 0.504–0.521 | 1 / 0 |
| 13 | [Kiro](https://feedbackbench.com/coding-agents/agents/kiro.md) | AWS | 41.6 | 40.4–43.1 | 12–15 | 0.404 | 0.478 | 0.467–0.490 | 0 / 4 |
| 14 | [Google Antigravity](https://feedbackbench.com/coding-agents/agents/antigravity.md) | Google | 41.3 | 40.0–42.7 | 13–15 | 0.807 | 0.394 | 0.382–0.405 | 2 / 30 |
| 15 | [Warp](https://feedbackbench.com/coding-agents/agents/warp.md) | Warp | 40.6 | 39.9–41.3 | 13–15 | 0.246 | 0.502 | 0.497–0.508 | 0 / 0 |
| 16 | [Grok Build](https://feedbackbench.com/coding-agents/agents/grok-build.md) | xAI | 35.6 | 35.1–36.2 | 16–16 | 0.030 | 0.507 | 0.502–0.512 | 0 / 0 |
| 17 | [Augment Code](https://feedbackbench.com/coding-agents/agents/augment.md) | Augment | 34.1 | 33.6–34.6 | 17–17 | 0.000 | 0.501 | 0.497–0.504 | 0 / 0 |

## Top quadrant

Rule: popularity of at least 0.5 and customer love of at least 0.5. At the edge: the customer love 95% interval still includes 0.5.

| Agent | Rank | Score | Popularity | Customer love | Customer love 95% interval | At the edge |
|---|---|---|---|---|---|---|
| Claude Code | 1 | 68.5 | 1.00 | 0.573 | 0.564–0.582 | no |
| Devin | 2 | 64.2 | 0.64 | 0.611 | 0.595–0.626 | no |
| OpenCode | 3 | 63.2 | 0.84 | 0.562 | 0.550–0.571 | no |
| Pi | 4 | 62.4 | 0.64 | 0.596 | 0.582–0.610 | no |
| Zed | 8 | 48.7 | 0.54 | 0.507 | 0.496–0.519 | yes |

## Week by week

Weekly rank and Feedback Score; each week is scored with the same method on that week's posts alone.

| Agent | Week of 2026-08-31 | Week of 2026-09-07 | Week of 2026-09-14 | Week of 2026-09-21 | Week of 2026-09-28 | Week of 2026-10-05 | Week of 2026-10-12 |
|---|---|---|---|---|---|---|---|
| Claude Code | rank 15 (13–15), score 39.9 | rank 8 (6–11), score 48.8 | rank 4 (4–7), score 55.6 | rank 1 (1–1), score 79.3 | rank 1 (1–1), score 78.2 | rank 1 (1–1), score 77.8 | rank 1 (1–2), score 70.2 |
| Devin | rank 5 (3–7), score 55.7 | rank 2 (2–2), score 67.7 | rank 2 (1–3), score 67.4 | rank 2 (2–3), score 61.8 | rank 4 (3–5), score 56.3 | rank 4 (3–5), score 56.8 | rank 3 (3–5), score 58.9 |
| OpenCode | rank 1 (1–1), score 76.3 | rank 1 (1–1), score 73.4 | rank 1 (1–2), score 69.4 | rank 3 (2–4), score 59.3 | rank 2 (2–3), score 62.1 | rank 2 (2–3), score 61.5 | rank 2 (1–2), score 68.5 |
| Pi | rank 3 (2–5), score 59.1 | rank 3 (3–5), score 59.5 | rank 3 (2–3), score 63.6 | rank 4 (3–5), score 56.3 | rank 3 (2–4), score 59.8 | rank 3 (2–4), score 58.4 | rank 6 (4–7), score 54.7 |
| Amp | rank 7 (5–10), score 51.6 | rank 7 (6–10), score 49.3 | rank 5 (4–9), score 53.4 | rank 6 (5–8), score 51.0 | rank 7 (6–11), score 49.7 | rank 5 (4–6), score 55.1 | rank 5 (3–6), score 55.9 |
| Factory | rank 10 (8–11), score 48.1 | rank 12 (10–14), score 45.0 | rank 7 (5–10), score 51.6 | rank 10 (7–11), score 48.2 | rank 6 (5–8), score 51.5 | rank 7 (6–8), score 52.3 | rank 4 (3–6), score 56.8 |
| Cursor | rank 4 (2–7), score 57.6 | rank 4 (3–5), score 59.2 | rank 6 (4–10), score 52.9 | rank 5 (4–7), score 52.7 | rank 5 (4–6), score 55.0 | rank 6 (4–7), score 53.9 | rank 7 (5–9), score 51.8 |
| Zed | rank 9 (7–11), score 49.5 | rank 10 (8–12), score 47.0 | rank 8 (5–11), score 51.3 | rank 8 (6–10), score 49.3 | rank 9 (7–11), score 48.7 | rank 9 (8–9), score 48.8 | rank 8 (7–9), score 50.4 |
| GitHub Copilot | rank 8 (5–11), score 51.3 | rank 6 (6–10), score 49.8 | rank 11 (6–12), score 49.4 | rank 7 (5–10), score 50.3 | rank 8 (6–11), score 49.1 | rank 8 (7–9), score 50.1 | rank 10 (9–12), score 47.7 |
| OpenAI Codex | rank 2 (2–5), score 59.4 | rank 9 (6–12), score 47.1 | rank 12 (11–13), score 44.9 | rank 13 (12–15), score 43.4 | rank 14 (13–15), score 42.0 | rank 12 (10–13), score 45.2 | rank 9 (7–11), score 49.6 |
| Cline | rank 11 (8–11), score 47.7 | rank 11 (8–14), score 45.9 | rank 10 (6–11), score 49.7 | rank 9 (6–11), score 48.8 | rank 10 (7–11), score 48.5 | rank 10 (10–12), score 45.8 | rank 11 (10–12), score 46.9 |
| Conductor | rank 12 (12–14), score 43.4 | rank 13 (10–14), score 44.8 | rank 13 (12–13), score 42.8 | rank 14 (13–14), score 43.1 | rank 12 (11–13), score 44.6 | rank 13 (13–13), score 43.2 | rank 13 (13–13), score 43.2 |
| Kiro | rank 14 (12–15), score 43.2 | rank 14 (11–15), score 44.2 | rank 14 (14–15), score 39.9 | rank 11 (10–12), score 46.5 | rank 13 (11–14), score 44.0 | rank 11 (10–12), score 45.5 | rank 12 (10–12), score 46.4 |
| Google Antigravity | rank 6 (3–9), score 54.6 | rank 5 (3–5), score 56.0 | rank 9 (5–12), score 49.9 | rank 12 (10–14), score 45.5 | rank 11 (8–13), score 46.5 | rank 15 (13–16), score 40.4 | rank 15 (15–16), score 37.1 |
| Warp | rank 13 (12–15), score 43.2 | rank 15 (14–15), score 42.2 | rank 15 (14–15), score 39.3 | rank 15 (14–15), score 41.8 | rank 15 (14–15), score 41.3 | rank 14 (14–15), score 41.8 | rank 14 (14–14), score 42.0 |
| Grok Build | rank 16 (16–16), score 37.0 | rank 16 (16–16), score 36.8 | rank 16 (16–16), score 35.0 | rank 16 (16–16), score 37.5 | rank 16 (16–16), score 37.9 | rank 17 (17–17), score 34.4 | rank 17 (17–17), score 34.6 |
| Augment Code | rank 17 (17–17), score 32.5 | rank 17 (17–17), score 33.5 | rank 17 (17–17), score 33.9 | rank 17 (17–17), score 35.2 | rank 17 (17–17), score 34.8 | rank 16 (15–16), score 38.9 | rank 16 (15–16), score 35.1 |

## Criteria

63 criteria, most discussed first, with the agent of highest customer love among agents with at least 30 rated user-weeks.

| Criterion | Area | Rated user-weeks | Highest customer love |
|---|---|---|---|
| [Can do your kind of task](https://feedbackbench.com/coding-agents/criteria/work.capability.md) | Quality of work | 15665 | Claude Code 0.698 |
| [Plan worth the price](https://feedbackbench.com/coding-agents/criteria/limits.plan_value.md) | Pricing and limits | 13894 | OpenCode 0.664 |
| [Usage quota lasts](https://feedbackbench.com/coding-agents/criteria/limits.burn_rate.md) | Pricing and limits | 13190 | Pi 0.639 |
| [Quality doesn't slip over time](https://feedbackbench.com/coding-agents/criteria/models.quality_drift.md) | Models | 6451 | Claude Code 0.693 |
| [Fast responses](https://feedbackbench.com/coding-agents/criteria/rel.response_speed.md) | Reliability | 5313 | Claude Code 0.684 |
| [Fair quota resets](https://feedbackbench.com/coding-agents/criteria/limits.reset_schedule.md) | Pricing and limits | 5196 | Claude Code 0.552 |
| [Subagents work well](https://feedbackbench.com/coding-agents/criteria/work.multi_agent_orchestration.md) | Quality of work | 4128 | Amp 0.556 |
| [Clear display and settings](https://feedbackbench.com/coding-agents/criteria/ui.display_settings.md) | Interface | 3749 | Claude Code 0.587 |
| [No cutoffs mid-task](https://feedbackbench.com/coding-agents/criteria/limits.window_interrupts_work.md) | Pricing and limits | 3494 | Claude Code 0.559 |
| [Runs without crashing](https://feedbackbench.com/coding-agents/criteria/rel.client_failures.md) | Reliability | 3491 | Claude Code 0.613 |
| [Plan terms don't change under you](https://feedbackbench.com/coding-agents/criteria/limits.allowance_change.md) | Pricing and limits | 3348 | OpenCode 0.618 |
| [Offers the models you want](https://feedbackbench.com/coding-agents/criteria/models.catalog_access.md) | Models | 3013 | OpenCode 0.645 |
| [Few outages and errors](https://feedbackbench.com/coding-agents/criteria/rel.service_errors.md) | Reliability | 2472 | Amp 0.514 |
| [Clear usage meter](https://feedbackbench.com/coding-agents/criteria/limits.usage_meter.md) | Pricing and limits | 2227 | Claude Code 0.641 |
| [MCP and plugins work](https://feedbackbench.com/coding-agents/criteria/setup.extensions_mcp.md) | Setup | 1989 | Pi 0.583 |
| [Follows instructions](https://feedbackbench.com/coding-agents/criteria/context.instruction_following.md) | Instructions and context | 1906 | OpenAI Codex 0.517 |
| [Helpful support](https://feedbackbench.com/coding-agents/criteria/account.support.md) | Support and account | 1774 | Amp 0.612 |
| [Auto mode picks the right model](https://feedbackbench.com/coding-agents/criteria/models.routing_auto.md) | Models | 1657 | Claude Code 0.574 |
| [Stays in scope](https://feedbackbench.com/coding-agents/criteria/work.scope_overreach.md) | Quality of work | 1539 | OpenCode 0.538 |
| [Easy install and sign-in](https://feedbackbench.com/coding-agents/criteria/setup.install_signin.md) | Setup | 1485 | Amp 0.540 |
| [Good mobile and remote control](https://feedbackbench.com/coding-agents/criteria/surfaces.remote_mobile.md) | Interface | 1420 | Amp 0.559 |
| [Compaction keeps what matters](https://feedbackbench.com/coding-agents/criteria/context.compaction.md) | Instructions and context | 1407 | Pi 0.543 |
| [Doesn't get stuck](https://feedbackbench.com/coding-agents/criteria/work.stuck_loops.md) | Quality of work | 1403 | OpenCode 0.566 |
| [Runs long unattended](https://feedbackbench.com/coding-agents/criteria/work.long_running_autonomy.md) | Quality of work | 1357 | Claude Code 0.557 |
| [Good free tier](https://feedbackbench.com/coding-agents/criteria/billing.free_tier.md) | Pricing and limits | 1328 | Devin 0.548 |
| [Effort setting works](https://feedbackbench.com/coding-agents/criteria/models.effort_control.md) | Models | 1314 | Claude Code 0.580 |
| [Follows project rules](https://feedbackbench.com/coding-agents/criteria/context.instruction_files.md) | Instructions and context | 1314 | Claude Code 0.535 |
| [Updates don't break things](https://feedbackbench.com/coding-agents/criteria/rel.update_breakage.md) | Reliability | 1222 | OpenCode 0.635 |
| [Sensible permission prompts](https://feedbackbench.com/coding-agents/criteria/work.permission_prompts.md) | Quality of work | 1216 | Claude Code 0.574 |
| [Remembers across sessions](https://feedbackbench.com/coding-agents/criteria/context.session_memory.md) | Instructions and context | 1185 | Claude Code 0.545 |
| [Good frontend UI](https://feedbackbench.com/coding-agents/criteria/work.frontend_ui.md) | Quality of work | 1109 | Claude Code 0.562 |
| [Concise, clear replies](https://feedbackbench.com/coding-agents/criteria/work.response_verbosity.md) | Quality of work | 1075 | OpenAI Codex 0.574 |
| [No false safety refusals](https://feedbackbench.com/coding-agents/criteria/work.safety_refusals.md) | Quality of work | 1074 | OpenAI Codex 0.636 |
| [Good IDE integration](https://feedbackbench.com/coding-agents/criteria/setup.ide_integration.md) | Setup | 1061 | Claude Code 0.587 |
| [Clear pricing](https://feedbackbench.com/coding-agents/criteria/billing.pricing_clarity.md) | Pricing and limits | 1055 | Claude Code 0.544 |
| [Stays sharp in long sessions](https://feedbackbench.com/coding-agents/criteria/context.long_context_decay.md) | Instructions and context | 1024 | OpenAI Codex 0.525 |
| [Cloud sessions work](https://feedbackbench.com/coding-agents/criteria/surfaces.cloud_sessions.md) | Interface | 1023 | Amp 0.590 |
| [No unfair bans](https://feedbackbench.com/coding-agents/criteria/account.bans_restrictions.md) | Support and account | 1019 | OpenAI Codex 0.552 |
| [Reliable session history](https://feedbackbench.com/coding-agents/criteria/ui.session_history.md) | Interface | 1016 | Pi 0.559 |
| [Code review finds real bugs](https://feedbackbench.com/coding-agents/criteria/verify.agent_code_review.md) | Checking the work | 952 | OpenAI Codex 0.585 |
| [Charges are correct](https://feedbackbench.com/coding-agents/criteria/account.billing_errors.md) | Support and account | 936 | Cursor 0.555 |
| [Own keys and local models work](https://feedbackbench.com/coding-agents/criteria/setup.provider_byok_local.md) | Setup | 933 | Pi 0.544 |
| [Your subscription works in other tools](https://feedbackbench.com/coding-agents/criteria/billing.subscription_portability.md) | Pricing and limits | 861 | OpenAI Codex 0.597 |
| [Good browser and computer use](https://feedbackbench.com/coding-agents/criteria/work.computer_browser_use.md) | Quality of work | 856 | OpenAI Codex 0.593 |
| [No risky actions unasked](https://feedbackbench.com/coding-agents/criteria/work.destructive_actions.md) | Quality of work | 846 | Claude Code 0.563 |
| [Prompt cache works](https://feedbackbench.com/coding-agents/criteria/limits.prompt_cache.md) | Pricing and limits | 827 | Pi 0.541 |
| [Doesn't break working code](https://feedbackbench.com/coding-agents/criteria/work.regressions_introduced.md) | Quality of work | 767 | Claude Code 0.526 |
| [Honest about 'done'](https://feedbackbench.com/coding-agents/criteria/verify.false_completion.md) | Checking the work | 738 | OpenAI Codex 0.565 |
| [Finds the right files](https://feedbackbench.com/coding-agents/criteria/context.codebase_retrieval.md) | Instructions and context | 661 | Cursor 0.545 |
| [Tests its own work](https://feedbackbench.com/coding-agents/criteria/verify.self_testing.md) | Checking the work | 646 | Claude Code 0.563 |
| [Easy to learn](https://feedbackbench.com/coding-agents/criteria/setup.onboarding_docs.md) | Setup | 582 | Pi 0.537 |
| [No surprise overage charges](https://feedbackbench.com/coding-agents/criteria/billing.overage_charges.md) | Pricing and limits | 574 | OpenAI Codex 0.556 |
| [Finishes the task](https://feedbackbench.com/coding-agents/criteria/work.premature_stop.md) | Quality of work | 553 | Claude Code 0.630 |
| [Finds and fixes bugs](https://feedbackbench.com/coding-agents/criteria/work.bug_diagnosis.md) | Quality of work | 515 | OpenCode 0.513 |
| [Useful plan mode](https://feedbackbench.com/coding-agents/criteria/work.plan_mode.md) | Quality of work | 512 | Claude Code 0.552 |
| [Protects your data](https://feedbackbench.com/coding-agents/criteria/account.data_privacy.md) | Support and account | 486 | Claude Code 0.502 |
| [Easy to review its changes](https://feedbackbench.com/coding-agents/criteria/verify.change_review_ui.md) | Checking the work | 439 | Cursor 0.567 |
| [Honest pushback](https://feedbackbench.com/coding-agents/criteria/work.sycophancy_pushback.md) | Quality of work | 385 | Claude Code 0.526 |
| [Handles git well](https://feedbackbench.com/coding-agents/criteria/work.git_workflow.md) | Quality of work | 376 | Claude Code 0.525 |
| [Asks when unsure](https://feedbackbench.com/coding-agents/criteria/context.clarifying_questions.md) | Instructions and context | 322 | Claude Code 0.500 |
| [Reads images and files](https://feedbackbench.com/coding-agents/criteria/context.attachments.md) | Instructions and context | 280 | Claude Code 0.552 |
| [Easy to interrupt and steer mid-run](https://feedbackbench.com/coding-agents/criteria/ui.interrupt_steer.md) | Interface | 257 | Cursor 0.534 |
| [Doesn't game the checks](https://feedbackbench.com/coding-agents/criteria/work.reward_hacking.md) | Quality of work | 160 | Google Antigravity 0.511 |

## Pages and data

- Agents: [Claude Code](https://feedbackbench.com/coding-agents/agents/claude-code.md), [Devin](https://feedbackbench.com/coding-agents/agents/devin.md), [OpenCode](https://feedbackbench.com/coding-agents/agents/opencode.md), [Pi](https://feedbackbench.com/coding-agents/agents/pi.md), [Amp](https://feedbackbench.com/coding-agents/agents/amp.md), [Factory](https://feedbackbench.com/coding-agents/agents/factory.md), [Cursor](https://feedbackbench.com/coding-agents/agents/cursor.md), [Zed](https://feedbackbench.com/coding-agents/agents/zed.md), [GitHub Copilot](https://feedbackbench.com/coding-agents/agents/copilot.md), [OpenAI Codex](https://feedbackbench.com/coding-agents/agents/codex.md), [Cline](https://feedbackbench.com/coding-agents/agents/cline.md), [Conductor](https://feedbackbench.com/coding-agents/agents/conductor.md), [Kiro](https://feedbackbench.com/coding-agents/agents/kiro.md), [Google Antigravity](https://feedbackbench.com/coding-agents/agents/antigravity.md), [Warp](https://feedbackbench.com/coding-agents/agents/warp.md), [Grok Build](https://feedbackbench.com/coding-agents/agents/grok-build.md), [Augment Code](https://feedbackbench.com/coding-agents/agents/augment.md)
- Criteria: one page per area and criterion under https://feedbackbench.com/coding-agents/criteria/ (list in [https://feedbackbench.com/coding-agents/llms.txt](https://feedbackbench.com/coding-agents/llms.txt))
- Articles, analysis separate from the data: [https://feedbackbench.com/coding-agents/articles.md](https://feedbackbench.com/coding-agents/articles.md)
- All data as JSON: [https://feedbackbench.com/coding-agents/data/coding.json](https://feedbackbench.com/coding-agents/data/coding.json)
- Ranking as CSV: [https://feedbackbench.com/coding-agents/data/rankings.csv](https://feedbackbench.com/coding-agents/data/rankings.csv)
- Every agent × criterion as CSV: [https://feedbackbench.com/coding-agents/data/criteria.csv](https://feedbackbench.com/coding-agents/data/criteria.csv)
- Top requests per agent and criterion as CSV: [https://feedbackbench.com/coding-agents/data/requests.csv](https://feedbackbench.com/coding-agents/data/requests.csv)