Gecko Tests · Knowledge HorizonLatest run 2026-11-16 · 55 models · 34 months · 272 questions

Knowledge Horizon · Where Does the Model’s World End?

Labs publish a training cutoff. We measure it: questions about each month's world events since January 2024, answered with no web access. The horizon is the last month the model still knows about.

In plain wordsEvery AI model stops learning at some date. We ask about real news, month by month, to find the last month each model actually knows about. A later month = more up to date.

ModelMeasured horizonClaimed cutoffReleasedAccuracy by month2024 accuracy
Claude Opus 5.5May 2026not published2026-09-22
99%
Claude Sonnet 5.5May 2026not published2026-09-28
99%
DeepSeek V4.1 Flash 1017May 2026not published2026-10-17
99%
DeepSeek V4.1 FlashMay 2026not published2026-09-10
99%
Qwen3.8 Max (0902)May 2026not published2026-09-03
98%
Qwen3.8 Max PrimeMay 2026not published2026-09-23
99%
MiniMax M3.1 FlashMay 2026not published2026-10-19
88%
GPT-6.1 Terra ProApr 2026not published2026-10-26
99%
GPT-6 SolApr 2026not published2026-09-22
100%
GPT-6 Luna ProApr 2026not published2026-09-22
98%
GPT-6 LunaApr 2026not published2026-09-22
99%
GPT-6.1 TerraApr 2026not published2026-10-26
99%
Kimi K3.1 ProApr 2026not published2026-10-23
98%
Qwen3.8 27BApr 2026not published2026-08-14
77%
Claude Opus 5Mar 2026not published2026-07-24
98%
GPT-6.1 Sol ProMar 2026not published2026-09-29
98%
GPT-6.1 SolMar 2026not published2026-09-29
98%
Qwen3.8 2.4T A95BMar 2026not published2026-08-12
98%
Claude Haiku 5.5Mar 2026not published2026-10-07
96%
Mistral Small 4Mar 2026not published2026-03-16
71%
Qwen4 MaxFeb 2026not published2026-11-16
99%
Gemini 4 HydrogenFeb 2026not published2026-11-12
99%
GLM 5.3 FlashXJan 2026not published2026-09-18
100%
GLM 5.3 FlashJan 2026not published2026-08-26
100%
GLM 5.3 FlashX LegalJan 2026not published2026-11-02
99%
GLM 5.3 PrimeDec 2025not published2026-09-23
99%
GLM 5.3Dec 2025not published2026-08-18
99%
DeepSeek V4 Pro 0813Nov 2025not published2026-08-12
100%
Kimi K3Nov 2025not published2026-07-16
99%
MiniMax M3Nov 2025not published2026-05-31
98%
Qwen3.8 FlashOct 2025not published2026-08-26
96%
Grok 4.7Oct 2025not published2026-09-21
98%
MiniMax M2-herOct 2025not published2026-01-23
64%
Gemini 3.6 FlashJul 2025not published2026-07-21
100%
DeepSeek V4 Flash 0731Jun 2025not published2026-07-31
99%
Kimi K2.7 CodeJun 2025not published2026-06-12
98%
DeepSeek V4 FlashApr 2025not published2026-04-24
98%
Mistral Large 4Apr 2025not published2026-10-06
99%
DeepSeek V4 ProApr 2025not published2026-04-24
98%
Kimi K2.5Apr 2025not published2026-01-27
94%
Gemini 3.7 FlashMar 2025not published2026-08-13
99%
Gemini 3.8 FlashMar 2025not published2026-09-02
99%
Nano Banana 2.1Feb 2025not published2026-10-06
97%
GLM 5.2Feb 2025not published2026-06-16
99%
Grok 4.5Feb 2025not published2026-07-08
99%
Gemini 4 ArgonFeb 2025not published2026-10-13
92%
Gemini 3.5 Flash LiteFeb 2025not published2026-07-21
99%
Grok Build 0.1Feb 2025not published2026-05-20
97%
Grok 4.6Jan 2025not published2026-08-12
98%
MiniMax M2.5Dec 2024not published2026-02-12
84%
Grok 4.3Dec 2024not published2026-04-30
91%
Mistral Medium 3.5Nov 2024not published2026-04-30
79%
Devstral 2 2512Nov 2024not published2025-12-09
79%
Kimi K2.6Nov 2024not published2026-04-20
83%
MiniMax M2.7Oct 2024not published2026-03-18
52%

Each bar is one month from Jan 2024; height = share answered correctly; pink = measured horizon.

How it works

  1. For each month, a writing model drafts questions from Wikipedia's Current events portal. A question is kept only if its answer appears word for word in the source event line and not in the question.
  2. Eight questions per month from January 2024. A new month is added automatically after it ends.
  3. Models answer with a short phrase or "unknown", temperature 0, lowest reasoning setting, no tools. Grading is string matching against the answer and its aliases.
  4. Horizon = the last month whose three-month rolling accuracy is at least half of the model's 2024 accuracy.
Open dataset · v1

272 items, every model answer and every score are public under CC BY 4.0.

Download JSON

How to cite · data as of 2026-11-16

BenchGecko Gecko Tests, Knowledge Horizon v1. BenchGecko, data as of 2026-11-16. https://benchgecko.ai/gecko-tests/knowledge-horizon

BenchGecko measures this data itself: free to reuse under CC BY 4.0 with the credit "Source: BenchGecko" and a link. JSON · llms.txt · MCP

Training data thins out in the final months before a cutoff, and some labs publish only a year. The measured horizon shows what the model can actually use.