{ "title": "LightRead 云端 Agent 缓存命中统计", "captured_at": "2026-09-30T22:53:12.267571+08:00", "window": { "start_inclusive": "2026-09-23T00:00:00+08:00", "end_exclusive": "2026-09-30T00:00:00+08:00", "timezone": "Asia/Shanghai" }, "source": "LightRead 生产用量记录的只读聚合；未包含对话内容、用户或任务标识", "scope": "窗口内发起的 cloud_chat 模型调用；平台模型；已结算、未标记中断、完整输入量大于零；未命中记录保留", "definitions": { "call_hit_rate": "cache_read_tokens / full_input_tokens", "weighted_hit_rate": "sum(cache_read_tokens) / sum(full_input_tokens)", "input_semantics": "按平台展示口径统一；Anthropic Messages 口径将普通输入、缓存读取、缓存写入相加，其余口径保留已含缓存的完整输入量", "task_groups": "按任务标识汇总窗口内调用，非完整生命周期统计或已完成任务数" }, "raw_window_record_count": 27379, "excluded_counts": { "marked_aborted": 140, "not_marked_finalized": 76, "byok": 1331, "no_input_tokens": 2 }, "calls": { "count": 25830, "input_tokens": 6658155565, "cache_read_tokens": 6065537514, "token_weighted_hit_rate": 0.9109936610500329, "median_hit_rate": 0.9907796014390147, "at_least_90_percent_count": 20961, "at_least_99_percent_count": 13307, "zero_hit_count": 422, "bins": { "below_50_percent": 2626, "50_to_below_90_percent": 2243, "90_to_below_99_percent": 7654, "at_least_99_percent": 13307 } }, "task_groups_within_window": { "count": 2115, "input_tokens": 6658155565, "cache_read_tokens": 6065537514, "token_weighted_hit_rate": 0.9109936610500329, "median_hit_rate": 0.9334914055899166, "at_least_90_percent_count": 1223, "at_least_99_percent_count": 487, "zero_hit_count": 59, "bins": { "below_50_percent": 206, "50_to_below_90_percent": 686, "90_to_below_99_percent": 736, "at_least_99_percent": 487 } }, "model_mix": [ { "model": "deepseek-v4.1-flash", "count": 18867, "input_tokens": 5173903752, "cache_read_tokens": 4862305657, "token_weighted_hit_rate": 0.9397750499553552, "median_hit_rate": 0.9939353537521998, "at_least_90_percent_count": 15890, "at_least_99_percent_count": 10965, "zero_hit_count": 8, "bins": { "below_50_percent": 1271, "50_to_below_90_percent": 1706, "90_to_below_99_percent": 4925, "at_least_99_percent": 10965 } }, { "model": "luna", "count": 2084, "input_tokens": 407952273, "cache_read_tokens": 267827584, "token_weighted_hit_rate": 0.6565169548644726, "median_hit_rate": 0.9531560514684831, "at_least_90_percent_count": 1243, "at_least_99_percent_count": 461, "zero_hit_count": 74, "bins": { "below_50_percent": 679, "50_to_below_90_percent": 162, "90_to_below_99_percent": 782, "at_least_99_percent": 461 } }, { "model": "sol", "count": 1669, "input_tokens": 246769951, "cache_read_tokens": 214766464, "token_weighted_hit_rate": 0.8703104374324733, "median_hit_rate": 0.9844305904625127, "at_least_90_percent_count": 1317, "at_least_99_percent_count": 649, "zero_hit_count": 58, "bins": { "below_50_percent": 214, "50_to_below_90_percent": 138, "90_to_below_99_percent": 668, "at_least_99_percent": 649 } }, { "model": "claude-sonnet-5", "count": 594, "input_tokens": 286847283, "cache_read_tokens": 266023570, "token_weighted_hit_rate": 0.9274048797596595, "median_hit_rate": 0.9936892847493703, "at_least_90_percent_count": 510, "at_least_99_percent_count": 367, "zero_hit_count": 39, "bins": { "below_50_percent": 50, "50_to_below_90_percent": 34, "90_to_below_99_percent": 143, "at_least_99_percent": 367 } }, { "model": "gemini-3.8-flash", "count": 573, "input_tokens": 125477630, "cache_read_tokens": 98601414, "token_weighted_hit_rate": 0.7858087055039213, "median_hit_rate": 0.9675757763590157, "at_least_90_percent_count": 425, "at_least_99_percent_count": 56, "zero_hit_count": 115, "bins": { "below_50_percent": 116, "50_to_below_90_percent": 32, "90_to_below_99_percent": 369, "at_least_99_percent": 56 } }, { "model": "astra", "count": 504, "input_tokens": 90770902, "cache_read_tokens": 61562752, "token_weighted_hit_rate": 0.6782212211574146, "median_hit_rate": 0.9627843519459451, "at_least_90_percent_count": 292, "at_least_99_percent_count": 162, "zero_hit_count": 29, "bins": { "below_50_percent": 159, "50_to_below_90_percent": 53, "90_to_below_99_percent": 130, "at_least_99_percent": 162 } }, { "model": "claude-opus-5.5", "count": 368, "input_tokens": 103571051, "cache_read_tokens": 94912011, "token_weighted_hit_rate": 0.9163951710792236, "median_hit_rate": 0.986437437721901, "at_least_90_percent_count": 312, "at_least_99_percent_count": 148, "zero_hit_count": 36, "bins": { "below_50_percent": 41, "50_to_below_90_percent": 15, "90_to_below_99_percent": 164, "at_least_99_percent": 148 } }, { "model": "mimo-v2.6-pro", "count": 312, "input_tokens": 60038832, "cache_read_tokens": 54114816, "token_weighted_hit_rate": 0.9013302590563388, "median_hit_rate": 0.9823741408367619, "at_least_90_percent_count": 246, "at_least_99_percent_count": 92, "zero_hit_count": 12, "bins": { "below_50_percent": 23, "50_to_below_90_percent": 43, "90_to_below_99_percent": 154, "at_least_99_percent": 92 } }, { "model": "glm-5.3", "count": 300, "input_tokens": 39090297, "cache_read_tokens": 35335335, "token_weighted_hit_rate": 0.903941328458057, "median_hit_rate": 0.9740753055996523, "at_least_90_percent_count": 238, "at_least_99_percent_count": 61, "zero_hit_count": 22, "bins": { "below_50_percent": 25, "50_to_below_90_percent": 37, "90_to_below_99_percent": 177, "at_least_99_percent": 61 } }, { "model": "claude-sonnet-5.5", "count": 271, "input_tokens": 93419036, "cache_read_tokens": 84676304, "token_weighted_hit_rate": 0.9064138062824797, "median_hit_rate": 0.9866255220074651, "at_least_90_percent_count": 232, "at_least_99_percent_count": 111, "zero_hit_count": 18, "bins": { "below_50_percent": 24, "50_to_below_90_percent": 15, "90_to_below_99_percent": 121, "at_least_99_percent": 111 } }, { "model": "grok-4.7", "count": 236, "input_tokens": 17925663, "cache_read_tokens": 17109184, "token_weighted_hit_rate": 0.9544519496991548, "median_hit_rate": 0.9954472582437531, "at_least_90_percent_count": 226, "at_least_99_percent_count": 220, "zero_hit_count": 0, "bins": { "below_50_percent": 9, "50_to_below_90_percent": 1, "90_to_below_99_percent": 6, "at_least_99_percent": 220 } }, { "model": "kimi-k3", "count": 26, "input_tokens": 2803284, "cache_read_tokens": 2400768, "token_weighted_hit_rate": 0.856412693112792, "median_hit_rate": 0.9918929899335094, "at_least_90_percent_count": 18, "at_least_99_percent_count": 14, "zero_hit_count": 4, "bins": { "below_50_percent": 5, "50_to_below_90_percent": 3, "90_to_below_99_percent": 4, "at_least_99_percent": 14 } }, { "model": "terra", "count": 16, "input_tokens": 8874245, "cache_read_tokens": 5550848, "token_weighted_hit_rate": 0.6255008735954439, "median_hit_rate": 0.9564302551313313, "at_least_90_percent_count": 10, "at_least_99_percent_count": 1, "zero_hit_count": 3, "bins": { "below_50_percent": 6, "50_to_below_90_percent": 0, "90_to_below_99_percent": 9, "at_least_99_percent": 1 } }, { "model": "mimo-v2.6-flash", "count": 7, "input_tokens": 563855, "cache_read_tokens": 259264, "token_weighted_hit_rate": 0.45980615583793705, "median_hit_rate": 0.666097463443103, "at_least_90_percent_count": 2, "at_least_99_percent_count": 0, "zero_hit_count": 3, "bins": { "below_50_percent": 3, "50_to_below_90_percent": 2, "90_to_below_99_percent": 2, "at_least_99_percent": 0 } }, { "model": "qwen3.8-max", "count": 3, "input_tokens": 147511, "cache_read_tokens": 91543, "token_weighted_hit_rate": 0.6205842276169234, "median_hit_rate": 0.8453847239300906, "at_least_90_percent_count": 0, "at_least_99_percent_count": 0, "zero_hit_count": 1, "bins": { "below_50_percent": 1, "50_to_below_90_percent": 2, "90_to_below_99_percent": 0, "at_least_99_percent": 0 } } ], "limitations": [ "该窗口为使用记录观察，不是控制实验或未来表现承诺。", "短任务、冷启动和未报告缓存的调用没有按低命中率排除。", "不同模型和任务的使用比例影响总体结果。", "缓存命中率不证明 API 来源，也不直接等于费用节省比例。", "总任务费用还包括输出、工具、算力和其他收费项目。" ], "arithmetic_cross_check": true }