Результаты привязаны к точному ID модели. Условия запуска и ревизия указаны первичным источником и могут отличаться от текущей ревизии.
overall-pass-rateagents-last-exam
20,4
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=148/152; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
overall-scoreagents-last-exam
42,9
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=148/152; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
ale-cli-pass-rateagents-last-exam
19,0
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=104/105; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
ale-cli-scoreagents-last-exam
43,5
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=104/105; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
near-term-pass-rateagents-last-exam
31,3
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=66/67; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
near-term-scoreagents-last-exam
60,0
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=66/67; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
full-spectrum-pass-rateagents-last-exam
21,8
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=55/55; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
full-spectrum-scoreagents-last-exam
42,5
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=55/55; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
last-exam-pass-rateagents-last-exam
2,63
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=35/38; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
last-exam-scoreagents-last-exam
14,7
результат · 01.10.2026
harness=claude_code; harness_variant=reasoning-max; attempted=35/38; official leaderboard snapshot=2026-10-01T20:39:17.826877+00:00; values are percentages (0-100); equal-task aggregation; unattempted tasks contribute zero; pass threshold score>=0.999. This is an official leaderboard export, not an HF-verified evaluation.
multimodalClaw-Eval
57,4
результат · 14.08.2026
Reported as ClawEval-MM. Pass@3, the benchmark's own Pass³ methodology; card also reports a secondary 56.9 'Average' metric, not included here.
deep_swedeep-swe
42,2
результат · 14.08.2026
meanExtractBench
89,8
результат · 24.08.2026
Pipeline name: qwen3_8_27b_fp8_vllm_extract_oneshot_structured_output_file; served checkpoint: Qwen/Qwen3.8-27B-FP8
shortExtractBench
94,7
результат · 24.08.2026
Pipeline name: qwen3_8_27b_fp8_vllm_extract_oneshot_structured_output_file; served checkpoint: Qwen/Qwen3.8-27B-FP8
mediumExtractBench
87,5
результат · 24.08.2026
Pipeline name: qwen3_8_27b_fp8_vllm_extract_oneshot_structured_output_file; served checkpoint: Qwen/Qwen3.8-27B-FP8
longExtractBench
38,5
результат · 24.08.2026
Pipeline name: qwen3_8_27b_fp8_vllm_extract_oneshot_structured_output_file; served checkpoint: Qwen/Qwen3.8-27B-FP8
Научное рассуждениеGPQA
89,2%№ 20 в рейтинге
точность · 14.08.2026
hlehle
30,8
результат · 14.08.2026
Judged by GPT-4o.
prompt_level_looseIFBench_test
79,5
результат · 14.08.2026
mmi-valuemmi
11,5
результат · 20.10.2026
modality-presence-scoremmi
18,1
результат · 20.10.2026
error-ratemmi
41,1
результат · 20.10.2026
pass-rate-strictmmi
9,18
результат · 20.10.2026
pass-rate-lenientmmi
13,3
результат · 20.10.2026
recallmmi
16,9
результат · 20.10.2026
precisionmmi
99,0
результат · 20.10.2026
OmniCompass-VLOmniCompass
29,6
результат · 26.10.2026
OmniCompass-VL; Overall CS on a 0-100 scale, higher is better. 32 video frames; paired Text and Text+Image; TG, SG and QA. Imported from the benchmark authors' published results, rounded to one decimal. Evaluation settings and model-specific limits follow the cited source. This submission does not claim HF verified status.
meanParseBench
70,8
результат · 28.08.2026
Pipeline name: qwen3_8_27b_thinking_parse_with_layout; served checkpoint: Qwen/Qwen3.8-27B-FP8
text_contentParseBench
88,3
результат · 28.08.2026
Pipeline name: qwen3_8_27b_thinking_parse_with_layout; served checkpoint: Qwen/Qwen3.8-27B-FP8
text_formattingParseBench
59,8
результат · 28.08.2026
Pipeline name: qwen3_8_27b_thinking_parse_with_layout; served checkpoint: Qwen/Qwen3.8-27B-FP8
layoutParseBench
69,9
результат · 28.08.2026
Pipeline name: qwen3_8_27b_thinking_parse_with_layout; served checkpoint: Qwen/Qwen3.8-27B-FP8
chartParseBench
69,2
результат · 28.08.2026
Pipeline name: qwen3_8_27b_thinking_parse_with_layout; served checkpoint: Qwen/Qwen3.8-27B-FP8
tableParseBench
66,8
результат · 28.08.2026
Pipeline name: qwen3_8_27b_thinking_parse_with_layout; served checkpoint: Qwen/Qwen3.8-27B-FP8
SWE_Bench_ProSWE-bench_Pro
61,7
результат · 14.08.2026
Evaluated with the Claude Code harness, temp=1.0, top_p=0.95, 256K context; baseline models re-evaluated on the same refined task set.
terminalbench_4terminal-bench
5,60
результат · 30.10.2026
Qwen3.8 27B (Xhigh); mini-swe-agent harness, pass@1 averaged over three repeats.
terminalbench_2_1terminal-bench-2.1
73
результат · 14.08.2026
Row labeled "(Terminus)" as the harness; no further hyperparameter footnote given for this row.
terminalbenchscience_0_1terminal-bench-science
0,50
результат · 30.10.2026
Qwen3.8 27B (Xhigh); mini-swe-agent harness, pass@1 averaged over three repeats.
overallWildClawBench
48,0
результат · 16.08.2026
avg_timeWildClawBench
516
результат · 16.08.2026
Оценки разных бенчмарков напрямую не сравниваются и не являются гарантией качества в вашем сценарии. Открыть все рейтинги →