Myth check: “AI progress has hit a wall”
Verdict: FalseAI kept improving fast through 2026: task length went from 3 hours (GPT-5, 2025) to about 40. Bigger models did slow, and real use lags.
METR time horizon: the length of software tasks (measured in skilled-human time) an AI agent completes with 50% or 80% success. Log scale.
What this measures: How big a job an AI agent can finish on its own. A task's "length" is how long it takes a skilled human; the chart shows the longest tasks an AI completes half the time (blue) or reliably (orange). Tasks are software, machine-learning and cybersecurity work.
Dashed line: exponential fit since 1 Jan 2023, doubling roughly every 4.0 months.
No longer updated · last data Apr 2026: METR stopped updating this measure on 8 September 2026. The newest models finish almost every task in its test suite, and METR says results above about 16 hours are unreliable, so its last data point is Claude Mythos Preview (April 2026), and no GPT-6, Claude 5.x or Gemini 4 model has a score. METR has moved to measuring how much AI speeds up AI research inside labs. Now watch: METR time horizon on the harder TH1.1 + TH2 suites (METR, 20 November 2026).
METR Time Horizon 1.1 values (after METR's March 2026 correction), plotted at model release dates. Confidence intervals are wide (e.g. Claude Opus 4.6: 317–3,634 min).
| Series | Date | Model | Value |
|---|---|---|---|
| 50% success | 14 Feb 2019 | GPT-2 | 6 s |
| 50% success | 28 May 2020 | davinci-002 (GPT-3) | 6 s |
| 50% success | 15 Mar 2022 | GPT-3.5 Turbo Instruct | 36 s |
| 50% success | 14 Mar 2023 | GPT-4 | 4 min |
| 50% success | 6 Nov 2023 | GPT-4 1106 | 4 min |
| 50% success | 4 Mar 2024 | Claude 3 Opus | 4 min |
| 50% success | 9 Apr 2024 | GPT-4 Turbo | 3.7 min |
| 50% success | 13 May 2024 | GPT-4o | 7 min |
| 50% success | 20 Jun 2024 | Claude 3.5 Sonnet (Jun) | 11 min |
| 50% success | 12 Sept 2024 | o1-preview | 20 min |
| 50% success | 22 Oct 2024 | Claude 3.5 Sonnet (Oct) | 21 min |
| 50% success | 5 Dec 2024 | o1 | 39 min |
| 50% success | 24 Feb 2025 | Claude 3.7 Sonnet | 1 h |
| 50% success | 16 Apr 2025 | o3 | 2 h |
| 50% success | 22 May 2025 | Claude Opus 4 | 1.7 h |
| 50% success | 5 Aug 2025 | Claude Opus 4.1 | 1.7 h |
| 50% success | 7 Aug 2025 | GPT-5 | 3.4 h |
| 50% success | 18 Nov 2025 | Gemini 3 Pro | 3.7 h |
| 50% success | 19 Nov 2025 | GPT-5.1-Codex-Max | 3.7 h |
| 50% success | 24 Nov 2025 | Claude Opus 4.5 | 4.9 h |
| 50% success | 11 Dec 2025 | GPT-5.2 | 5.9 h |
| 50% success | 5 Feb 2026 | Claude Opus 4.6 | 12 h |
| 50% success | 5 Feb 2026 | GPT-5.3-Codex | 5.8 h |
| 50% success | 19 Feb 2026 | Gemini 3.1 Pro | 6.4 h |
| 50% success | 5 Mar 2026 | GPT-5.4 | 5.7 h |
| 50% success | 7 Apr 2026 | Claude Mythos Preview (limited access) | 17 h |
| 80% success | 15 Mar 2022 | GPT-3.5 Turbo Instruct | 18 s |
| 80% success | 14 Mar 2023 | GPT-4 | 54 s |
| 80% success | 6 Nov 2023 | GPT-4 1106 | 48 s |
| 80% success | 13 May 2024 | GPT-4o | 1.3 min |
| 80% success | 20 Jun 2024 | Claude 3.5 Sonnet (Jun) | 1.7 min |
| 80% success | 12 Sept 2024 | o1-preview | 4.4 min |
| 80% success | 5 Dec 2024 | o1 | 7.1 min |
| 80% success | 24 Feb 2025 | Claude 3.7 Sonnet | 12 min |
| 80% success | 16 Apr 2025 | o3 | 30 min |
| 80% success | 7 Aug 2025 | GPT-5 | 38 min |
| 80% success | 18 Nov 2025 | Gemini 3 Pro | 54 min |
| 80% success | 11 Dec 2025 | GPT-5.2 | 1.1 h |
| 80% success | 5 Feb 2026 | Claude Opus 4.6 | 1.2 h |
| 80% success | 19 Feb 2026 | Gemini 3.1 Pro | 1.5 h |
| 80% success | 7 Apr 2026 | Claude Mythos Preview (limited access) | 3.1 h |
AI kept improving fast through 2026: task length went from 3 hours (GPT-5, 2025) to about 40. Bigger models did slow, and real use lags.
In 2023 the best AI could do, half the time, software tasks taking a skilled person about 4 minutes. By February 2026 it was about 12 hours.
4 min → 12 htasks AI finishes half the time, GPT-4 (2023) → Claude Opus 4.6 (2026)
METR stopped updating the old measure in September 2026, then replaced it in November with harder tests.
~12 hold METR tests, best public AI, Feb–Mar 2026 (harder new suite: ~40 h)