ME News reports that on April 16 (UTC+8), according to monitoring by Beating, the AI safety evaluation firm METR updated its "Time Horizon" benchmark, incorporating new test data for Google’s Gemini 3.1 Pro. This benchmark tracks the upper limit of autonomous programming task completion by advanced AI agents and has become a key reference for measuring AI agent capability growth since its launch in February this year. The measurement involves having human software engineering experts (averaging about five years of experience) and AI agents complete the same set of over 100 software tasks, with human completion time used to gauge task difficulty. The two core metrics are the 50% Time Horizon (the highest task difficulty at which the AI has a 50% probability of success) and the 80% Time Horizon (the highest task difficulty at which the AI has an 80% probability of success). Gemini 3.1 Pro reversed its ranking between these two metrics. At the 50% Time Horizon, it ranked second, behind the significantly leading Claude Opus 4.6: 1. Claude Opus 4.6: ~12.0 hours 2. Gemini 3.1 Pro: ~6.4 hours 3. GPT-5.2: ~5.9 hours 4. GPT-5.4: ~5.7 hours However, under the stricter 80% Time Horizon, Gemini 3.1 Pro surpassed all others to take first place: 1. Gemini 3.1 Pro: ~1.5 hours 2. Claude Opus 4.6: ~1.2 hours 3. GPT-5.2: ~1.1 hours Claude Opus 4.6 can tackle more difficult tasks but exhibits higher variability in success rates, while Gemini 3.1 Pro has a slightly lower ceiling but demonstrates greater consistency within its capability range—making it potentially more practical for production environments requiring predictable outcomes. Compared to its predecessor, Gemini 3 Pro (~3.7 hours at the 50% Time Horizon), Gemini 3.1 Pro shows a ~71% improvement. Looking at the longer-term trend, METR’s data shows that the time horizon for leading models has grown from a few seconds in 2019 (GPT-2) to over ten hours today, roughly doubling every 4.3 months—METR notes, “We see no signs of slowing in this exponential growth.” Note that METR’s tasks cover software engineering, machine learning, and cybersecurity—all clearly defined, independently scorable tasks. In follow-up research, METR found that when scoring shifted from algorithmic evaluation to human holistic assessment, AI performance dropped significantly. A 12-hour time horizon does not equate to an AI being able to replace half a day of human work in practice. (Source: BlockBeats)
Gemini 3.1 Pro Outperforms All Leading AI Models in Reliability on METR's Time Horizon Benchmark
KuCoinFlashShare
Summary
On-chain news: On April 16 (UTC+8), METR updated its 'Time Horizon' benchmark with data for Gemini 3.1 Pro. The model ranked second in the 50% Time Horizon (6.4 hours) and first in the 80% Time Horizon (1.5 hours), outperforming GPT-5.2 and Claude Opus 4.6. METR highlighted Gemini 3.1 Pro’s enhanced stability for production use. Compared to Gemini 3 Pro, it demonstrates a 71% improvement in the 50% Time Horizon. AI + crypto news continues to underscore key advancements in AI performance and reliability.
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information.
Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.