BenchLM
An independent AI model evaluation aggregation platform that consolidates multi-dimensional benchmark results and provides composite scores and cross-model comparisons.
Core Facts
Timeline (last 90 days)
BenchLM's methodology page mentions that BenchAlign will "calibrate benchmark protocols, human preference signals, and external composite indices" before combining them, but it does not list the specific names of the external indices, so it cannot be determined whether the agreement between the two leaderboards includes a circular-validation component.
On BenchLM, Kimi K3 scores 69.92 and ranks 18th overall (out of 226 models), but it does not appear in the open-weight list because its weight access is marked as "pending" and there is "no verified weight release".
BenchLM lists Qwen3.8 Max as second among open-weight models, but Artificial Analysis marks it as isOpenWeights: false; on the Hugging Face Hub API, anonymous requests to Qwen/Qwen3.8-Max and Qwen/Qwen3.8-Max-0902 both return HTTP 401.
Data from both BenchLM and Artificial Analysis on October 19 show MiMo-V2.6-Pro ranked first among open-weight models, with BenchLM giving 73.7 points and Artificial Analysis giving an intelligence index of 46.3.
BenchLM marks the row containing MiMo-V2.6-Pro as Estimated and gives a conditional interval of 63.92 to 83.37, an interval width of 19.45 points; rows marked as Supported in the same table have interval widths of only 8.2 to 10.0 points.
Among the top nine models in BenchLM's open-weight ranking, all models' conditional intervals overlap each other, with none of the 36 pairs as an exception; after adding the tenth row, all 45 pairs also overlap.
Matching BenchLM's top ten open-weight rows against Artificial Analysis by model name, nine models corresponded successfully (dots3-note Preview is missing from the latter's data); ranking these nine models separately gives a Kendall's tau of 0.61, with 29 concordant pairs and 7 discordant pairs.
In the cross-domain tests of the World of AI Benchmark, DeepSeek V4.1 Flash 1017 ranks 10th to 11th overall.
According to Anthropic's system card, Claude Opus 5.5 reaches 89.9% accuracy on SWE-bench Pro, and the BenchLM October 2026 leaderboard also repeats this figure.
All Facts (9)
In the cross-domain tests of the World of AI Benchmark, DeepSeek V4.1 Flash 1017 ranks 10th to 11th overall.
65%UnverifiedData from both BenchLM and Artificial Analysis on October 19 show MiMo-V2.6-Pro ranked first among open-weight models, with BenchLM giving 73.7 points and Artificial Analysis giving an intelligence index of 46.3.
50%UnverifiedBenchLM marks the row containing MiMo-V2.6-Pro as Estimated and gives a conditional interval of 63.92 to 83.37, an interval width of 19.45 points; rows marked as Supported in the same table have interval widths of only 8.2 to 10.0 points.
50%UnverifiedAmong the top nine models in BenchLM's open-weight ranking, all models' conditional intervals overlap each other, with none of the 36 pairs as an exception; after adding the tenth row, all 45 pairs also overlap.
50%UnverifiedMatching BenchLM's top ten open-weight rows against Artificial Analysis by model name, nine models corresponded successfully (dots3-note Preview is missing from the latter's data); ranking these nine models separately gives a Kendall's tau of 0.61, with 29 concordant pairs and 7 discordant pairs.
50%UnverifiedOn BenchLM, Kimi K3 scores 69.92 and ranks 18th overall (out of 226 models), but it does not appear in the open-weight list because its weight access is marked as "pending" and there is "no verified weight release".
50%UnverifiedBenchLM lists Qwen3.8 Max as second among open-weight models, but Artificial Analysis marks it as isOpenWeights: false; on the Hugging Face Hub API, anonymous requests to Qwen/Qwen3.8-Max and Qwen/Qwen3.8-Max-0902 both return HTTP 401.
50%UnverifiedBenchLM's methodology page mentions that BenchAlign will "calibrate benchmark protocols, human preference signals, and external composite indices" before combining them, but it does not list the specific names of the external indices, so it cannot be determined whether the agreement between the two leaderboards includes a circular-validation component.
50%UnverifiedAccording to Anthropic's system card, Claude Opus 5.5 reaches 89.9% accuracy on SWE-bench Pro, and the BenchLM October 2026 leaderboard also repeats this figure.
50%