English
Subscribe

Measured Autonomy & Target Tier Matrix · No Affiliate Links · As of 2026-11-12

AI 5-Level Autonomy (L1–L5) & Target Positioning Tiers (Tier 1–4)

Company scale (ARR or MAU) alone does not tell you what a tool can actually execute. Compare AI products across two independent axes: (1) Autonomy Level (L1 Q&A to L5 Multi-Agent Infra) and (2) Target & Pricing Tier (Tier 1 Enterprise to Tier 4 Open-Weight).

1. The 5 Levels of AI Autonomy (Level 1 to Level 5)

Higher autonomy is not automatically better for every workflow. Moving from L2 to L3–L4 increases permission blocks and false completion claims, requiring automated verification harnesses.

Level 1

Conversational Q&A

Single prompt in, single text/summary out — no tool execution or multi-step state

Core Benchmark StandardMMLU-Pro · LMSYS Chatbot Arena · GPQA Diamond
Metering & Cost UnitFree to $20/mo flat ($0.15–$15 per 1M tokens)
Primary Failure ModeUnverified hallucinations · stale pricing or legal facts
Level 2

Copilot & Creative Studio

Human-in-the-loop studio generating shots, voices, or code snippets with immediate human review

Core Benchmark StandardShot acceptance rate · TTS transcript CER · credit cost per rendered minute
Metering & Cost Unit$10–$99/mo credit plans (metered by shot, character, or minute)
Primary Failure ModeCredits burned on failed takes · character drift · strict rollover expiration
Level 3

Personal Assistant & Workflow Agent

Connects calendar, reminders, messaging, web search, and memory to execute daily routines

Core Benchmark StandardAllPass Core 10-job measured pass rate · GAIA · scheduling & call accuracy
Metering & Cost UnitFree to $20–$30/mo subscription (or platform bundle)
Primary Failure ModeBlocked third-party permissions · false completion claims · unintended outbound actions
Level 4

Autonomous Coding & Local Execution Agent

Runs terminal, filesystem, browser, and test loops autonomously across multi-file repositories

Core Benchmark StandardSWE-bench Verified · Terminal-Bench · OSWorld · harness verification pass rate
Metering & Cost Unit$20–$200/mo seat + API token or ACU consumption
Primary Failure ModeClaiming completion without running tests · overwriting shared files · runaway token burn
Level 5

Enterprise & Sovereign Multi-Agent Infra

Operates across ERP/CRM, local GPU clusters, and MCP protocols with audit logs and RBAC

Core Benchmark StandardCost per resolution · SLA uptime · on-prem data sovereignty · security audit compliance
Metering & Cost Unit$0.10–$2 per resolution, enterprise annual contract, or self-hosted GPU
Primary Failure ModePrompt injection across connectors · permission leakage · pilot cancellation from unclear ROI

2. Target & Pricing Positioning Tiers (Tier 1 to Tier 4)

T1 Tier 1 · Enterprise & Custom

Annual contracts, dedicated SLA/security, outcome-based or custom volume pricing

T2 Tier 2 · Pro & Heavy Seat ($100–$200+/mo)

High-volume seats and studio plans for full-time engineers and production teams

T3 Tier 3 · Creator & Prosumer ($10–$49/mo)

Standard monthly subscriptions used by solo founders, creators, and practitioners

T4 Tier 4 · Mass, Free & Open-Weight ($0+)

Free tiers and open-weight local models (Apache 2.0 / MIT) with $0 subscription cost

3. Interactive 2-Axis Matrix (Autonomy Level × Target Tier)

Autonomy Level
Target Tier
Category
Tool / ModelAutonomyTarget TierScale TierOfficial PricingMeasured & Key MetricCards & Maps
Claude Code
Coding & CLI
L4T2APro $20 · Max $100/$200 · API usageTerminal multi-file autonomous coding · used daily by our desk ($2.5B+ run-rate)
OpenAI Codex
Coding & CLI
L4T3BIncluded in Plus $20 / Pro $200 · open-source CLICloud sandbox parallel tasks + local CLI cross-check (5M+ WAU)
Pickle
Personal Agent
L3T3CBasic $20 · Pro $50/mo10-job measured test: 6 done · 3 partly · 1 blocked (placed 1 real family call)
ChatGPT · Dots
Personal Agent
L3T3SFree to Plus $20 · Pro $200/mo10-job measured test: 5 done · 2 partly · 3 declined (strong web research)
Grok Bot
Personal Agent
L3T3AX Premium · SuperGrok $30/mo10-job measured test: 4 done · 3 partly · 3 declined (real-time X pulse)
Google Flow (Veo)
Video Studio
L2T2SGoogle AI Pro $19.99 · Ultra $249.99/moPrimary video generator across 95 shorts by our desk · reference-locked shots
Higgsfield
Video Studio
L2T3BBasic $9 · Pro $29 · Ultimate $49/moCamera motion presets and stylized shot control for video creators
ElevenLabs
Voice & TTS
L2T3BFree to Starter $5 · Creator $22 · Pro $99/moMultilingual TTS, dubbing, and ElevenAgents (15M weekly conversations)
Typecast
Voice & TTS
L2T3CFree to Basic · Pro · Business monthly plansStudio TTS specialized in Korean emotional nuance and character acting
Qwen3-TTS (Local Open-Weight)
Voice & TTS
L3T4BFree open-weight (Apache 2.0 · local GPU compute only)Powers ainewground's 4-language article audio narration on local GPU · $0/mo fee
OpenAI Whisper (Local ASR Audit)
ASR & Audit
L3T4AFree open-source (MIT · local) / API $0.006/minCore ASR engine for automated TTS-to-script harness verification after synthesis
Salesforce Agentforce & MCP Infra
Enterprise & Infra
L5T1A$125–$550/seat/mo + $0.10/action / MCP open standardAgentforce ARR $1.5B+ (2026-07) · MCP SDK 97M monthly downloads

4. What Our 10-Job Measured Tests Reveal About Autonomy Levels

  • Conversational fluency (L1) and end-to-end action execution (L3–L4) diverge sharply. In our 10-job test, Pickle completed 6 jobs and placed 1 real phone call, while third-party permissions blocked 1 job.
  • Creative studios (L2: Google Flow, ElevenLabs, Typecast) deduct credits even on discarded takes. Pairing them with local ASR verification (Whisper L3) keeps per-video costs predictable.
  • Instead of paying Tier 2 ($100–$200/mo) across the board, combine Tier 2 autonomous coding (Claude Code) with Tier 4 open-weight local models (Qwen3-TTS) to cut recurring overhead.

See Pickle vs ChatGPT Dots vs Grok Bot Measured Comparison →Explore All 20 Sector Market Maps (with 2-Axis Toggle) →