# Feedback Bench 1.1 > Personal agents ranked by what their users say in public (Reddit, X, YouTube, App Store, Play Store). 13 agents, 53 criteria in 9 areas, built 2026-11-16, window 2026-08-31 to 2026-11-15. Every number comes from labelled public posts; every page lists its posts with links. The web page renders with JavaScript. Every view below is also available as static Markdown and every number as JSON or CSV, so agents do not need to run the page. 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). In the data files, popularity is `reach` and customer love is `regard`. ## Start here - [Ranking, top quadrant, week by week and criteria](https://feedbackbench.com/index.md): the landing page as Markdown - [Method](https://feedbackbench.com/method.md): sources, labelling, formulas, parameters, weights, and all 53 criterion definitions - [Everything in one file](https://feedbackbench.com/llms-full.txt): ranking, method, and every agent page without posts ## Agents - [Muse](https://feedbackbench.com/agents/muse.md): rank 1, Feedback Score 76.6, Meta - [Hermes Agent](https://feedbackbench.com/agents/hermes-agent.md): rank 2, Feedback Score 63.9, Nous Research - [Hark](https://feedbackbench.com/agents/hark.md): rank 3, Feedback Score 55.4, Hark Labs - [Grok Bot](https://feedbackbench.com/agents/grok-bot.md): rank 4, Feedback Score 51.1, xAI - [OpenClaw](https://feedbackbench.com/agents/openclaw.md): rank 5, Feedback Score 50.9, OpenClaw Foundation - [Claude Duet](https://feedbackbench.com/agents/duet.md): rank 6, Feedback Score 47.9, Anthropic - [Instinct](https://feedbackbench.com/agents/instinct.md): rank 7, Feedback Score 47.2, Spear Street Technology - [Wabi](https://feedbackbench.com/agents/wabi.md): rank 8, Feedback Score 46.9, Wabi - [Gemini Spark](https://feedbackbench.com/agents/gemini-spark.md): rank 9, Feedback Score 44.5, Google - [Fo](https://feedbackbench.com/agents/fo.md): rank 10, Feedback Score 43.6, Wajo - [Cue](https://feedbackbench.com/agents/cue.md): rank 11, Feedback Score 42.2, Manus - [ChatGPT Dots](https://feedbackbench.com/agents/dots.md): rank 12, Feedback Score 41.6, OpenAI - [Poke](https://feedbackbench.com/agents/poke.md): rank 13, Feedback Score 38.2, The Interaction Company ## Areas and criteria - [Paying for the agent](https://feedbackbench.com/criteria/billing.md): What does it cost me to use it, and is it worth it? - [Subscription price versus value received](https://feedbackbench.com/criteria/billing.price_level.md) - [Free tier, trial or required plan to start](https://feedbackbench.com/criteria/billing.entry_plan.md) - [Usage allowance size, exhaustion and reset wait](https://feedbackbench.com/criteria/billing.usage_cap.md) - [Agent spends usage on overhead and costly choices](https://feedbackbench.com/criteria/billing.token_efficiency.md) - [Plan contents, bundles and usage meters unclear](https://feedbackbench.com/criteria/billing.plan_clarity.md) - [Buying, upgrading, cancelling and refunding go wrong](https://feedbackbench.com/criteria/billing.purchase_flow.md) - [Getting access and set up](https://feedbackbench.com/criteria/access.md): Can I get the agent running on my devices and account? - [Country availability, waitlists and staged rollout](https://feedbackbench.com/criteria/access.availability.md) - [Apps offered for phone, tablet and desktop](https://feedbackbench.com/criteria/access.platform.md) - [Account creation, sign-in and identity checks fail](https://feedbackbench.com/criteria/access.sign_in.md) - [Effort needed to set up and train the agent](https://feedbackbench.com/criteria/access.setup.md) - [Connecting accounts, sites and devices](https://feedbackbench.com/criteria/connect.md): Can the agent reach the services, websites, devices and files it needs? - [Which services have a connector or integration](https://feedbackbench.com/criteria/connect.coverage.md) - [Linking accounts and staying logged in](https://feedbackbench.com/criteria/connect.account_link.md) - [Websites block the agent with captchas or IP bans](https://feedbackbench.com/criteria/connect.site_blocking.md) - [Access to the user's own phone or computer](https://feedbackbench.com/criteria/connect.local_device.md) - [Moving files between agent, apps and user](https://feedbackbench.com/criteria/connect.file_transfer.md) - [Talking to the agent](https://feedbackbench.com/criteria/ask.md): How easy and pleasant is it to give the agent requests and get replies? - [Reaching the agent through messaging apps](https://feedbackbench.com/criteria/ask.channel.md) - [Speaking with the agent by voice or call](https://feedbackbench.com/criteria/ask.voice.md) - [Threads, chats and parallel tasks organisation](https://feedbackbench.com/criteria/ask.conversation.md) - [Response and task completion speed](https://feedbackbench.com/criteria/ask.speed.md) - [App screens, layout, rendering, localisation and stability](https://feedbackbench.com/criteria/ask.app_client.md) - [Avatar, personality and tone of replies](https://feedbackbench.com/criteria/ask.persona.md) - [Getting tasks done](https://feedbackbench.com/criteria/act.md): Does the agent actually do the jobs I hand it, correctly and completely? - [Email triage, cleanup and drafting replies](https://feedbackbench.com/criteria/act.email.md) - [Calendar entries, conflicts and schedule changes](https://feedbackbench.com/criteria/act.calendar.md) - [Booking appointments, reservations, travel and tickets](https://feedbackbench.com/criteria/act.reservations.md) - [Buying products, orders and purchase judgment](https://feedbackbench.com/criteria/act.shopping.md) - [Forms, applications, claims and life admin paperwork](https://feedbackbench.com/criteria/act.admin.md) - [Budgets, transactions, bills and investments](https://feedbackbench.com/criteria/act.finance.md) - [Web research, monitoring and summarising information](https://feedbackbench.com/criteria/act.research.md) - [Writing code and building apps or websites](https://feedbackbench.com/criteria/act.build.md) - [Creating writing, images, video and social posts](https://feedbackbench.com/criteria/act.content.md) - [Contacting other people and businesses for you](https://feedbackbench.com/criteria/act.outreach.md) - [Operating a browser or desktop by clicking](https://feedbackbench.com/criteria/act.computer_use.md) - [Stops short, or hands the task back](https://feedbackbench.com/criteria/act.stops_short.md) - [Says it did something it did not do](https://feedbackbench.com/criteria/act.false_reports.md) - [Wrong facts, hallucinations and careless mistakes](https://feedbackbench.com/criteria/act.accuracy.md) - [Following instructions, rules and corrections](https://feedbackbench.com/criteria/act.instructions.md) - [Running on its own](https://feedbackbench.com/criteria/auto.md): Does the agent keep working, remembering and reminding without me? - [Unprompted suggestions, insights and check-ins](https://feedbackbench.com/criteria/auto.proactive.md) - [Scheduled recurring tasks and daily briefings run](https://feedbackbench.com/criteria/auto.routines.md) - [Reminders and notifications reach the user correctly](https://feedbackbench.com/criteria/auto.reminders.md) - [Multiple agents, teams and delegation between them](https://feedbackbench.com/criteria/auto.multi_agent.md) - [Where the agent runs: its own computer and always-on running](https://feedbackbench.com/criteria/auto.cloud_env.md) - [Remembering the user and context across sessions](https://feedbackbench.com/criteria/auto.memory.md) - [Staying in control](https://feedbackbench.com/criteria/control.md): Can I decide, see and shape what the agent does on my behalf? - [Asks before acting: too little or too much](https://feedbackbench.com/criteria/control.approval.md) - [Seeing what the agent is doing and has done](https://feedbackbench.com/criteria/control.visibility.md) - [Which AI model runs, how good it is, and choosing it](https://feedbackbench.com/criteria/control.model.md) - [Configuring behaviour, skills and portability](https://feedbackbench.com/criteria/control.customize.md) - [Trusting it with data and safety](https://feedbackbench.com/criteria/trust.md): Is my data and my life safe with this agent? - [How the vendor collects, keeps and uses data](https://feedbackbench.com/criteria/trust.data_use.md) - [Permission scope requested and isolation between agents](https://feedbackbench.com/criteria/trust.access_scope.md) - [Vulnerabilities, prompt injection and cross-user leaks](https://feedbackbench.com/criteria/trust.security.md) - [Refuses legitimate requests because of guardrails](https://feedbackbench.com/criteria/trust.refusals.md) - [Dealing with the vendor](https://feedbackbench.com/criteria/vendor.md): Does the company keep the agent running, improving and supported? - [Support responses and account bans or lockouts](https://feedbackbench.com/criteria/vendor.support.md) - [Outages, errors and agent going unresponsive](https://feedbackbench.com/criteria/vendor.uptime.md) - [Got better or worse over time, and update cadence](https://feedbackbench.com/criteria/vendor.updates.md) ## Articles Analysis, signed and dated, separate from the benchmark data. - [After Opus 5.5, Claude Code users still complain about the length or clarity of its replies at three times the field](https://feedbackbench.com/coding-agents/articles/after-opus-5-5-claude-code-users-still-complain-about-the-length-or-clarity-of-its-replies-at-three-times-the-field.md): Coding agents, Claude Code, 16 November 2026 - [OpenClaw users complain about setup at more than twice the rate of Hermes Agent users](https://feedbackbench.com/articles/openclaw-users-complain-about-setup-at-more-than-twice-the-rate-of-hermes-agent-users.md): Personal agents, OpenClaw, 15 November 2026 - [Codex is 1st of 6 on code review, the reviewer Claude Code users call in](https://feedbackbench.com/coding-agents/articles/codex-is-1st-of-6-on-code-review-the-reviewer-claude-code-users-call-in.md): Coding agents, OpenAI Codex, 14 November 2026 - [Grok Bot's positive share fell from 65% to 40% in one week, and the usage cap explains only part of it](https://feedbackbench.com/articles/grok-bots-positive-share-fell-from-65-to-40-in-one-week-and-the-usage-cap-explains-only-part-of-it.md): Personal agents, Grok Bot, 12 November 2026 - [Grok Bot users made it a dispatcher for Claude Code and Codex before xAI's routing](https://feedbackbench.com/articles/grok-bot-users-made-it-a-dispatcher-for-claude-code-and-codex-before-xais-routing.md): Personal agents, Grok Bot, 17 October 2026 - [Hermes Agent users choose their models, and Grok Bot users ask to do the same](https://feedbackbench.com/articles/hermes-agent-users-choose-their-models-and-grok-bot-users-ask-to-do-the-same.md): Personal agents, Hermes Agent, 17 October 2026 - [Gemini Spark users trust it to file their records; it pauses before deleting](https://feedbackbench.com/articles/gemini-spark-users-trust-it-to-file-their-records-it-pauses-before-deleting.md): Personal agents, Gemini Spark, 17 October 2026 - [Users fault personal agents more for holding back than for overstepping](https://feedbackbench.com/articles/users-fault-personal-agents-more-for-holding-back-than-for-overstepping.md): Personal agents, 17 October 2026 - [A personal agent's false “done” surfaces when someone looks, or when it costs](https://feedbackbench.com/articles/a-personal-agents-false-done-surfaces-when-someone-looks-or-when-it-costs.md): Personal agents, 17 October 2026 - [Zed users praise its speed more often than users of any other coding agent](https://feedbackbench.com/coding-agents/articles/zed-users-praise-its-speed-more-often-than-users-of-any-other-coding-agent.md): Coding agents, Zed, 15 October 2026 - [OpenCode is 1st of 12 on plan value, as the $10 worker beside a frontier model](https://feedbackbench.com/coding-agents/articles/opencode-is-1st-of-12-on-plan-value-as-the-10-worker-beside-a-frontier-model.md): Coding agents, OpenCode, 15 October 2026 - [Amp's users praise its cloud machines at a higher rate than any rival's](https://feedbackbench.com/coding-agents/articles/amps-users-praise-its-cloud-machines-at-a-higher-rate-than-any-rivals.md): Coding agents, Amp, 15 October 2026 - [Pi's users ask it to extend and repair itself, and it can read its own manual](https://feedbackbench.com/coding-agents/articles/pis-users-ask-it-to-extend-and-repair-itself-and-it-can-read-its-own-manual.md): Coding agents, Pi, 15 October 2026 - [Cursor is 1st of 5 on letting developers see and approve every change an agent makes](https://feedbackbench.com/coding-agents/articles/cursor-is-1st-of-5-on-letting-developers-see-and-approve-every-change-an-agent-makes.md): Coding agents, Cursor, 15 October 2026 - [A coding agent should not be the only judge of its own “done”](https://feedbackbench.com/coding-agents/articles/a-coding-agent-should-not-be-the-only-judge-of-its-own-done.md): Coding agents, 15 October 2026 ## Other categories - [Coding agents](https://feedbackbench.com/coding-agents/llms.txt): the same benchmark for coding agents ## Data - [personal.json](https://feedbackbench.com/data/personal.json): the full build: meta (rules, parameters, weights, baselines, criteria) and every agent with every criterion, intervals and posts - [rankings.csv](https://feedbackbench.com/data/rankings.csv): one row per agent - [criteria.csv](https://feedbackbench.com/data/criteria.csv): one row per agent and area or criterion - [requests.csv](https://feedbackbench.com/data/requests.csv): the top requests of every agent, area and criterion, counted in user-weeks ## Optional - [Web page](https://feedbackbench.com/): the interactive version for people - [Enterpret](https://www.enterpret.com): builds Feedback Bench; runs the same feedback analysis on a company's own customer feedback