| Anti-AI Tricks | A test is fully passed only if every run passed for that test. No failed answers. Response Time (avg) 3496ms Response Time (max) 4305ms Response Time (total) 10487ms A test is fully passed only if every run passed for that test. Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 100.0% Google: Gemini 3 Pro Preview - Reasoning (medium) 100.0% Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 100.0% Google: Gemini 3 Flash Preview - Reasoning (low) 100.0% MoonshotAI: Kimi K2.5 - No Reasoning 0.0% 0.0% 100.0% | 10.00 Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance. Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Pro Preview - Reasoning (medium) 10.00 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 10.00 Z.ai: GLM 4.7 Flash - No Reasoning 1.00 1.00 10.00 | 10.00 Consistency score reflects repeat-to-repeat stability (10 = very consistent, even if consistently wrong). Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Pro Preview - Reasoning (medium) 10.00 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 10.00 Anthropic: Claude Opus 4.6 - Reasoning (medium) 4.41 4.41 10.00 | 100.0% Attempt pass rate = passed attempts / total attempts across repeats. Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 100.0% Google: Gemini 3 Pro Preview - Reasoning (medium) 100.0% Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 100.0% Google: Gemini 3 Flash Preview - Reasoning (low) 100.0% xAI: Grok 4.1 Fast - No Reasoning 0.0% 0.0% 100.0% | 0 Flaky tests had mixed outcomes across repeats (at least one pass and one fail). Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 0 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 0 Google: Gemini 3 Pro Preview - Reasoning (medium) 0 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 0 Google: Gemini 3 Flash Preview - Reasoning (low) 0 Anthropic: Claude Opus 4.6 - Reasoning (medium) 2 0 2 | 6.23 Measures reasoning clarity, efficiency, and consistency independent of final answer correctness. Note: for some Gemini models, only partial reasoning text is available, so reasoning score may be understated. Rank: #14/19 28% Anthropic: Claude Opus 4.6 - Reasoning (medium) 10.00 OpenAI: gpt-oss-120b - Reasoning (medium) 10.00 Anthropic: Claude Sonnet 4.6 - Reasoning (medium) 9.89 Z.ai: GLM 5 - Reasoning (medium) 9.83 StepFun: Step 3.5 Flash - Reasoning (medium) 9.83 Google: Gemini 3 Flash Preview - Reasoning (low) 6.23 Qwen: Qwen3 Coder Next - Reasoning (medium) 4.00 4.00 10.00 | 3496ms | $0.00844 Total Cost Rank: #16/29 46% StepFun: Step 3.5 Flash - Reasoning (medium) $0.00000 OpenAI: GPT-4o-mini - No Reasoning $0.00018 Z.ai: GLM 4.7 Flash - No Reasoning $0.00020 Xiaomi: MiMo-V2-Flash - No Reasoning $0.00024 xAI: Grok 4.1 Fast - No Reasoning $0.00049 Google: Gemini 3 Flash Preview - Reasoning (low) $0.00844 Anthropic: Claude Opus 4.6 - Reasoning (medium) $0.05049 $0.00000 $0.05049 |
| Data parsing and extraction | A test is fully passed only if every run passed for that test. No failed answers. Response Time (avg) 9460ms Response Time (max) 14717ms Response Time (total) 18919ms A test is fully passed only if every run passed for that test. Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 100.0% Google: Gemini 3 Pro Preview - Reasoning (medium) 100.0% Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 100.0% Google: Gemini 3 Flash Preview - Reasoning (low) 100.0% Z.ai: GLM 4.7 Flash - No Reasoning 0.0% 0.0% 100.0% | 10.00 Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance. Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Pro Preview - Reasoning (medium) 10.00 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 10.00 Z.ai: GLM 4.7 Flash - No Reasoning 0.50 0.50 10.00 | 10.00 Consistency score reflects repeat-to-repeat stability (10 = very consistent, even if consistently wrong). Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Pro Preview - Reasoning (medium) 10.00 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 10.00 Z.ai: GLM 5 - Reasoning (medium) 5.56 5.56 10.00 | 100.0% Attempt pass rate = passed attempts / total attempts across repeats. Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 100.0% Google: Gemini 3 Pro Preview - Reasoning (medium) 100.0% Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 100.0% Google: Gemini 3 Flash Preview - Reasoning (low) 100.0% Xiaomi: MiMo-V2-Flash - No Reasoning 16.7% 0.0% 100.0% | 0 Flaky tests had mixed outcomes across repeats (at least one pass and one fail). Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 0 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 0 Google: Gemini 3 Pro Preview - Reasoning (medium) 0 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 0 Google: Gemini 3 Flash Preview - Reasoning (low) 0 Z.ai: GLM 5 - Reasoning (medium) 1 0 1 | 4.73 Measures reasoning clarity, efficiency, and consistency independent of final answer correctness. Note: for some Gemini models, only partial reasoning text is available, so reasoning score may be understated. Rank: #17/19 11% OpenAI: gpt-oss-120b - Reasoning (medium) 10.00 Z.ai: GLM 4.7 Flash - Reasoning (medium) 9.87 Anthropic: Claude Sonnet 4.6 - Reasoning (medium) 9.83 Anthropic: Claude Opus 4.6 - Reasoning (medium) 9.83 Z.ai: GLM 5 - Reasoning (medium) 9.80 Google: Gemini 3 Flash Preview - Reasoning (low) 4.73 OpenAI: GPT-5.3-Codex - Reasoning (medium) 1.25 1.25 10.00 | 9460ms | $0.01354 Total Cost Rank: #18/29 39% StepFun: Step 3.5 Flash - Reasoning (medium) $0.00000 Xiaomi: MiMo-V2-Flash - Reasoning (medium) $0.00029 Xiaomi: MiMo-V2-Flash - No Reasoning $0.00029 Z.ai: GLM 4.7 Flash - No Reasoning $0.00050 OpenAI: gpt-oss-120b - Reasoning (medium) $0.00052 Google: Gemini 3 Flash Preview - Reasoning (low) $0.01354 Anthropic: Claude Opus 4.6 - Reasoning (medium) $0.07755 $0.00000 $0.07755 |
| Domain specific | A test is fully passed only if every run passed for that test. Wrong answer: 2 Response Time (avg) 8314ms Response Time (max) 14399ms Response Time (total) 24941ms A test is fully passed only if every run passed for that test. Rank: #8/29 75% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 66.7% Google: Gemini 3 Flash Preview - No Reasoning 66.7% Anthropic: Claude Sonnet 4.6 - No Reasoning 66.7% Z.ai: GLM 4.7 Flash - No Reasoning 66.7% Google: Gemini 3 Flash Preview - Reasoning (low) 33.3% Anthropic: Claude Sonnet 4.6 - Reasoning (medium) 0.0% 0.0% 100.0% | 4.00 Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance. Rank: #8/29 75% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 7.00 Google: Gemini 3 Flash Preview - No Reasoning 7.00 Anthropic: Claude Sonnet 4.6 - No Reasoning 7.00 Z.ai: GLM 4.7 Flash - No Reasoning 7.00 Google: Gemini 3 Flash Preview - Reasoning (low) 4.00 Anthropic: Claude Sonnet 4.6 - Reasoning (medium) 1.00 1.00 10.00 | 4.41 Consistency score reflects repeat-to-repeat stability (10 = very consistent, even if consistently wrong). Rank: #23/29 21% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - No Reasoning 10.00 Anthropic: Claude Sonnet 4.6 - No Reasoning 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 4.41 4.41 10.00 | 55.5% Attempt pass rate = passed attempts / total attempts across repeats. Rank: #11/29 64% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 66.7% Google: Gemini 3 Flash Preview - No Reasoning 66.7% Anthropic: Claude Sonnet 4.6 - No Reasoning 66.7% Z.ai: GLM 4.7 Flash - No Reasoning 66.7% Google: Gemini 3 Flash Preview - Reasoning (low) 55.5% Z.ai: GLM 5 - No Reasoning 0.0% 0.0% 100.0% | 2 Flaky tests had mixed outcomes across repeats (at least one pass and one fail). Rank: #23/29 21% Google: Gemini 3 Flash Preview - Reasoning (medium) 0 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 0 Google: Gemini 3 Pro Preview - Reasoning (medium) 0 Google: Gemini 3 Flash Preview - No Reasoning 0 Anthropic: Claude Sonnet 4.6 - No Reasoning 0 Google: Gemini 3 Flash Preview - Reasoning (low) 2 0 2 | 1.83 Measures reasoning clarity, efficiency, and consistency independent of final answer correctness. Note: for some Gemini models, only partial reasoning text is available, so reasoning score may be understated. Rank: #18/19 6% Xiaomi: MiMo-V2-Flash - Reasoning (medium) 8.72 OpenAI: gpt-oss-120b - Reasoning (medium) 8.53 StepFun: Step 3.5 Flash - Reasoning (medium) 8.44 Z.ai: GLM 5 - Reasoning (medium) 8.43 Z.ai: GLM 4.7 Flash - Reasoning (medium) 8.21 Google: Gemini 3 Flash Preview - Reasoning (low) 1.83 OpenAI: GPT-5.3-Codex - Reasoning (medium) 1.00 1.00 8.72 | 8314ms | $0.01993 Total Cost Rank: #18/29 39% StepFun: Step 3.5 Flash - Reasoning (medium) $0.00000 Z.ai: GLM 4.7 Flash - No Reasoning $0.00005 Xiaomi: MiMo-V2-Flash - No Reasoning $0.00008 Qwen: Qwen3 Coder Next - No Reasoning $0.00010 Qwen: Qwen3 Coder Next - Reasoning (medium) $0.00010 Google: Gemini 3 Flash Preview - Reasoning (low) $0.01993 Anthropic: Claude Sonnet 4.6 - Reasoning (medium) $0.64205 $0.00000 $0.64205 |
| Instructions following | A test is fully passed only if every run passed for that test. Did not follow instructions: 1 Response Time (avg) 7016ms Response Time (max) 7350ms Response Time (total) 14031ms A test is fully passed only if every run passed for that test. Rank: #14/29 54% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 100.0% Google: Gemini 3 Pro Preview - Reasoning (medium) 100.0% Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 100.0% OpenAI: GPT-5.2 - Reasoning (medium) 100.0% Google: Gemini 3 Flash Preview - Reasoning (low) 50.0% xAI: Grok 4.1 Fast - No Reasoning 0.0% 0.0% 100.0% | 7.50 Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance. Rank: #15/29 50% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Anthropic: Claude Sonnet 4.6 - Reasoning (medium) 10.00 Z.ai: GLM 5 - No Reasoning 10.00 OpenAI: gpt-oss-120b - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 7.50 xAI: Grok 4.1 Fast - No Reasoning 1.00 1.00 10.00 | 9.99 Consistency score reflects repeat-to-repeat stability (10 = very consistent, even if consistently wrong). Rank: #17/29 43% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Pro Preview - Reasoning (medium) 10.00 OpenAI: GPT-5.3-Codex - Reasoning (medium) 10.00 OpenAI: GPT-5.2 - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 9.99 Xiaomi: MiMo-V2-Flash - Reasoning (medium) 5.80 5.80 10.00 | 50.0% Attempt pass rate = passed attempts / total attempts across repeats. Rank: #19/29 36% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 100.0% Google: Gemini 3 Pro Preview - Reasoning (medium) 100.0% Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 100.0% OpenAI: GPT-5.2 - Reasoning (medium) 100.0% Google: Gemini 3 Flash Preview - Reasoning (low) 50.0% xAI: Grok 4.1 Fast - No Reasoning 0.0% 0.0% 100.0% | 0 Flaky tests had mixed outcomes across repeats (at least one pass and one fail). Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 0 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 0 Google: Gemini 3 Pro Preview - Reasoning (medium) 0 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 0 Google: Gemini 3 Flash Preview - Reasoning (low) 0 Google: Gemini 3 Flash Preview - No Reasoning 1 0 1 | 5.00 Measures reasoning clarity, efficiency, and consistency independent of final answer correctness. Note: for some Gemini models, only partial reasoning text is available, so reasoning score may be understated. Rank: #17/19 11% Anthropic: Claude Sonnet 4.6 - Reasoning (medium) 10.00 Z.ai: GLM 5 - Reasoning (medium) 9.75 StepFun: Step 3.5 Flash - Reasoning (medium) 9.67 Anthropic: Claude Opus 4.6 - Reasoning (medium) 9.50 OpenAI: gpt-oss-120b - Reasoning (medium) 9.50 Google: Gemini 3 Flash Preview - Reasoning (low) 5.00 OpenAI: GPT-5.3-Codex - Reasoning (medium) 1.00 1.00 10.00 | 7016ms | $0.00878 Total Cost Rank: #20/29 32% StepFun: Step 3.5 Flash - Reasoning (medium) $0.00000 Z.ai: GLM 4.7 Flash - No Reasoning $0.00006 Xiaomi: MiMo-V2-Flash - No Reasoning $0.00008 Qwen: Qwen3 Coder Next - No Reasoning $0.00013 Qwen: Qwen3 Coder Next - Reasoning (medium) $0.00014 Google: Gemini 3 Flash Preview - Reasoning (low) $0.00878 Google: Gemini 3.1 Pro Preview - Reasoning (medium) $0.03134 $0.00000 $0.03134 |
| Puzzle Solving | A test is fully passed only if every run passed for that test. No failed answers. Response Time (avg) 6440ms Response Time (max) 10274ms Response Time (total) 19319ms A test is fully passed only if every run passed for that test. Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 100.0% Google: Gemini 3 Pro Preview - Reasoning (medium) 100.0% Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 100.0% Google: Gemini 3 Flash Preview - Reasoning (low) 100.0% StepFun: Step 3.5 Flash - Reasoning (medium) 0.0% 0.0% 100.0% | 10.00 Summarizes broad quality across our full private benchmark suite, so ranking reflects consistent performance. Rank: #4/29 89% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Pro Preview - Reasoning (medium) 10.00 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 10.00 Anthropic: Claude Sonnet 4.6 - Reasoning (medium) 10.00 Xiaomi: MiMo-V2-Flash - Reasoning (medium) 1.00 1.00 10.00 | 10.00 Consistency score reflects repeat-to-repeat stability (10 = very consistent, even if consistently wrong). Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 10.00 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 10.00 Google: Gemini 3 Pro Preview - Reasoning (medium) 10.00 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 10.00 Google: Gemini 3 Flash Preview - Reasoning (low) 10.00 MiniMax: MiniMax M2.5 - Reasoning (medium) 4.79 4.79 10.00 | 100.0% Attempt pass rate = passed attempts / total attempts across repeats. Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 100.0% Google: Gemini 3.1 Pro Preview - Reasoning (medium) 100.0% Google: Gemini 3 Pro Preview - Reasoning (medium) 100.0% Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 100.0% Google: Gemini 3 Flash Preview - Reasoning (low) 100.0% OpenAI: GPT-4o-mini - No Reasoning 0.0% 0.0% 100.0% | 0 Flaky tests had mixed outcomes across repeats (at least one pass and one fail). Rank: #5/29 86% Google: Gemini 3 Flash Preview - Reasoning (medium) 0 Google: Gemini 3.1 Pro Preview - Reasoning (medium) 0 Google: Gemini 3 Pro Preview - Reasoning (medium) 0 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) 0 Google: Gemini 3 Flash Preview - Reasoning (low) 0 OpenAI: GPT-5 Nano - Reasoning (medium) 2 0 2 | 7.50 Measures reasoning clarity, efficiency, and consistency independent of final answer correctness. Note: for some Gemini models, only partial reasoning text is available, so reasoning score may be understated. Rank: #13/19 33% Z.ai: GLM 5 - Reasoning (medium) 9.50 Anthropic: Claude Sonnet 4.6 - Reasoning (medium) 9.44 Anthropic: Claude Opus 4.6 - Reasoning (medium) 9.44 MoonshotAI: Kimi K2.5 - Reasoning (medium) 9.26 StepFun: Step 3.5 Flash - Reasoning (medium) 9.22 Google: Gemini 3 Flash Preview - Reasoning (low) 7.50 Qwen: Qwen3 Coder Next - Reasoning (medium) 4.33 4.33 9.50 | 6440ms | $0.01105 Total Cost Rank: #17/29 43% StepFun: Step 3.5 Flash - Reasoning (medium) $0.00000 Z.ai: GLM 4.7 Flash - No Reasoning $0.00008 OpenAI: GPT-4o-mini - No Reasoning $0.00028 xAI: Grok 4.1 Fast - No Reasoning $0.00053 Qwen: Qwen3 Coder Next - Reasoning (medium) $0.00058 Google: Gemini 3 Flash Preview - Reasoning (low) $0.01105 Qwen: Qwen3.5 Plus 2026-02-15 - Reasoning (medium) $0.05508 $0.00000 $0.05508 |