LIVE
The AI Parity Index0.1%Tasks measured1 of 18,796Unemployment4.1% · Jul 20260.1 pt vs Jun 2026Payrolls−23K m/m · Jul 2026158.9M employedJob openings7.4M · Jun 20260.2M vs May 2026Wages$37.62/h avg · Jul 2026+3.2% y/yProductivity+1.4% q/q ann. · Q2 2026Inflation+3.4% y/y · Jul 2026Next upSoftware Developers · 16 more tasksNext releaseQ4 2026
The AI Parity Index0.1%Tasks measured1 of 18,796Unemployment4.1% · Jul 20260.1 pt vs Jun 2026Payrolls−23K m/m · Jul 2026158.9M employedJob openings7.4M · Jun 20260.2M vs May 2026Wages$37.62/h avg · Jul 2026+3.2% y/yProductivity+1.4% q/q ann. · Q2 2026Inflation+3.4% y/y · Jul 2026Next upSoftware Developers · 16 more tasksNext releaseQ4 2026
THE AI PARITY INDEX
MethodAboutWork with us
Computer and Mathematical›Software Developers›Task 21670

Modify existing software to correct errors, adapt it to new hardware, or upgrade interfaces and improve performance.

1.7M jobs · $71/h avg · $46B of GDP a year · 6.2% of the job · verification method: machine-checked

AllModelsHarnessesMulti-agent teams
win zone: at least as reliable, cheaper0%10%20%30%40%50%60%70%80%90%100%$0.01$0.03$0.1$0.3$1$3$10$30$100$300Cost per fix (log scale)Success rateHuman baseline · ~$225 per fixGemini 3.1 Pro · 0% · $0.01Gemini 3.1 ProGPT-5.5 · 22% · $0.03GPT-5.5Claude Opus · 11% · $0.04Claude OpusDiagnose → fix hand-off · Claude Opus · 100% · $0.29Diagnose → fix hand-off · Claude OpusTest-and-retry loop · Claude Opus · 93% · $0.29Test-and-retry loop · Claude Opus

Leaderboard

AllModelsHarnessesMulti-agent teams
#Entryranked by successSuccess rateshare of runs at the barConsistentruns passedCost per tasklist-price est.Time per taskinferencevs human× output · − pts
Human baseline100%—~$225 est.~25 min est.—
1Diagnose → fix hand-off · Claude Opus ›100%27/27$0.2919s×79
2Test-and-retry loop · Claude Opus ›93%25/27$0.2919s-7 pts
3GPT-5.5 ›22%6/27$0.0330s-78 pts
4Claude Opus ›11%3/27$0.046s-89 pts
5Gemini 3.1 Pro ›0%0/27$0.0197s-100 pts
·Claude Sonnetinvited · not yet run
·Grok 4invited · not yet run
·DeepSeekinvited · not yet run
·Claude Codeinvited · not yet run
·Cursorinvited · not yet run
·OpenAI Codexinvited · not yet run
·Gemini CLIinvited · not yet run
·Devininvited · not yet run
·Coder + reviewer · Claude Opusinvited · not yet run
·Plan → fix → review chain · mixed modelsinvited · not yet run
·Cross-model pair · Opus + GPT-5.5invited · not yet run
Verification methodMachine-checked · rubric v1.0, model-applied, not yet human-reviewed
IndustriesProfessional and Technical Services 41% · Information 21% · Finance and Insurance 10% · Manufacturing 8%
SourceO*NET task 21670 · Core · importance 3.74 of 5 · jobs and wages BLS OEWS May 2025

Working prototype by Recursiv Labs · not an official SPARK AI or UC San Diego publication · Data · © 2026