{"schema_version": "research-pack/1", "object": {"id": "inference_cost", "title": "推理成本", "one_liner": "推理 $/百万 token 的成本曲线，agent 放量的前提"}, "facts": [{"valid_from": null, "valid_to": null, "published_at": null, "collected_at": null, "verified_at": null, "conditions": null, "conversion": null, "id": "fact:13d77316e69db9791274516a", "metric_id": "inference_cost", "label": "推理成本", "value": 0.1, "value_type": "number", "unit": "$/百万 token", "raw_value": 0.1, "raw_unit": "$/百万 token", "observed_at": "2026-09-10", "source_ids": ["src:93e9b5da4d436a97dbc7e9af"], "evidence_role": "unknown", "freshness": "unknown", "slug": "inference_cost", "title": "推理成本", "at": "2026-09-10", "src": "https://artificialanalysis.ai/models/glm-5-3-flash", "delta": {"dir": "flat", "abs": 0.0, "pct": 0.0}, "note": "GLM-5.3-Flash；AA 公示官方 API 输入 0.15 / 输出 0.50 美元每百万 token，7:2:1 混合 0.10。河级口径取独立评测站混合价，不替代 zhipu_flagship_aa。上一行 note 被 shell 把美元符展开，以本行为准。", "role": "home", "pending": false, "display_text": "- inference_cost｜推理成本｜0.1 $/百万 token @ 2026-09-10｜GLM-5.3-Flash；AA 公示官方 API 输入 0.15 / 输出 0.50 美元每百万 token，7:2:1 混合 0.10。河级口径取独立评测站混合价，不替代 zhipu_flagship_aa。上一行 note 被 shell 把美元符展开，以本行为准。｜主页｜https://artificialanalysis.ai/models/glm-5-3-flash"}], "timeline": [], "who": [{"id": "who:maxforai", "person_id": "maxforai", "name": "Max For AI", "at": "2026-10-16T03:32:33.000Z", "text": "- Max For AI｜人｜Max For AI：埃森哲自10月1日起将Claude Enterprise中央预算限为每月29美元，重度员工月耗500～1000美元｜2026-10-16｜https://x.com/MaxForAI/status/2110936912521560180｜/p/maxforai", "source_ids": ["src:3fa164d071059062e29dacaf"]}, {"id": "who:tp_huang", "person_id": "tp_huang", "name": "tphuang", "at": "2026-10-15T20:32:53.000Z", "text": "- tphuang｜人｜tphuang 称腾讯云应把部分算力让给 DeepSeek，因 DS V4.1F 服务成本远低于其他模型｜2026-10-15｜https://x.com/tphuang/status/2110831297207865435｜/p/tp_huang", "source_ids": ["src:a1c7bf4f93d13cb93bd0f44e"]}, {"id": "who:teortaxes", "person_id": "teortaxes", "name": "Teortaxes", "at": "2026-10-15T20:26:01.000Z", "text": "- Teortaxes｜人｜Teortaxes：DS V4.1F 推理成本远低于同类模型，DeepSeek 因不愿为推理投入过多算力而故意给 token 定高价｜2026-10-15｜https://x.com/teortaxesTex/status/2110829570052051256｜/p/teortaxes", "source_ids": ["src:da221b57ea0dd73a6389b4c4"]}, {"id": "who:prime_intellect", "person_id": "prime_intellect", "name": "Prime Intellect", "at": "2026-10-15T17:07:41.000Z", "text": "- Prime Intellect｜云｜Prime Intellect 称 Sword Health 用 prime-rl 将 GLM-5.3 后训练为心理健康模型 Dawn 1，更安全、更少谄媚、推理成本更低｜2026-10-15｜https://x.com/PrimeIntellect/status/2110779658723840142｜/o/prime_intellect", "source_ids": ["src:5822c3679fd6be8a133a5fb2"]}, {"id": "who:neal_k_patel", "person_id": "neal_k_patel", "name": "Neal Patel", "at": "2026-10-14T14:28:47.000Z", "text": "- Neal Patel｜人｜Neal Patel 称 ScaleDown 在分类/抽取/摘要任务上快 200 倍、便宜 467 倍｜2026-10-14｜https://x.com/neal_k_patel/status/2110377281570886097｜/p/neal_k_patel", "source_ids": ["src:815d2124b15bbdbf69ce6449"]}, {"id": "who:jerry_liu", "person_id": "jerry_liu", "name": "Jerry Liu", "at": "2026-10-09T17:38:21.000Z", "text": "- Jerry Liu｜人｜Jerry Liu 称不靠推理量赚差价，只抽小额交易分成并推动推理成本下降｜2026-10-09｜https://x.com/jerryjliu0/status/2108613049448108093｜/p/jerry_liu", "source_ids": ["src:d7992386c67832a8e0d9ed0b"]}], "context": ["> 谁在说 · 时间线为订户内容（https://hourahead.ai/enter）"], "usage": "只报告改变了已记录值的条目；每条给出处、时间、说话人；\n不得生成买卖建议、目标价、持仓建议；\n「到期仍待核对」不得读成「未发生」；厂商自测不得当第三方。", "issues": [{"id": "wp_222fa887a02f", "question": "Kimi 推理成本相对 DeepSeek 是否已脱离贵档", "status": "unknown", "window": "架构切换落地观测", "claim_ids": ["claim:1a7372850b1c8fb43768b542", "claim:d24650eb207f7bed0a7adc5f"], "source_ids": ["src:76523c5546c9d6efa7abdb06", "src:e29a0c313a4df1577d14d405"], "evidence_state": "sourced"}, {"id": "wp_b56a29829397", "question": "同等能力的最低推理成本是否仍每季降 47%", "status": "unknown", "window": "下次成本曲线观测", "claim_ids": ["claim:04278348ae9817a1e2ded468"], "source_ids": ["src:85989eab4e3d04afe401b4e4"], "evidence_state": "sourced"}, {"id": "wp_043f335dfe20", "question": "DS V4.1F 的 API 标价是否仍高于服务成本", "status": "unknown", "window": "下次服务成本披露", "claim_ids": ["claim:ecaf548fdd92bd966f462b8c"], "source_ids": ["src:7232eae48574fa341c6fa7be", "src:f2bf0214344d0400cd97526f"], "evidence_state": "sourced"}], "claims": [{"id": "claim:1a7372850b1c8fb43768b542", "issue_id": "wp_222fa887a02f", "person_id": "teortaxes", "who": "Teortaxes", "text": "10/17：可转 DSA+ 砍成本，但仍近不了 V4.1", "said_at": "10月17日", "source_ids": ["src:e29a0c313a4df1577d14d405"], "evidence_state": "sourced"}, {"id": "claim:d24650eb207f7bed0a7adc5f", "issue_id": "wp_222fa887a02f", "person_id": "teortaxes", "who": "Teortaxes", "text": "10/10：轻量 K2.8 若是 K2+，贵过 4 月后任何 DS", "said_at": "10月10日", "source_ids": ["src:76523c5546c9d6efa7abdb06"], "evidence_state": "sourced"}, {"id": "claim:04278348ae9817a1e2ded468", "issue_id": "wp_b56a29829397", "person_id": "maxforai", "who": "Max For AI", "text": "转述：2023 起同等能力最低成本每季降 47%", "said_at": "9月24日", "source_ids": ["src:85989eab4e3d04afe401b4e4"], "evidence_state": "sourced"}, {"id": "claim:ecaf548fdd92bd966f462b8c", "issue_id": "wp_043f335dfe20", "person_id": "teortaxes", "who": "Teortaxes", "text": "服务成本远低于同类，且故意把 token 定高价", "said_at": "10月16日", "source_ids": ["src:7232eae48574fa341c6fa7be"], "evidence_state": "sourced"}], "relations": [], "sources": [{"id": "src:93e9b5da4d436a97dbc7e9af", "url": "https://artificialanalysis.ai/models/glm-5-3-flash", "who": null, "person_id": null, "published_at": null, "evidence_role": "unknown", "event_id": null, "item_id": null, "source_role": "home", "source_tier": ""}, {"id": "src:e29a0c313a4df1577d14d405", "url": "https://x.com/teortaxesTex/status/2111427146182787407", "who": "Teortaxes", "person_id": "teortaxes", "published_at": "10月17日", "evidence_role": "unknown", "event_id": null, "item_id": "48002", "source_role": "", "source_tier": ""}, {"id": "src:76523c5546c9d6efa7abdb06", "url": "https://x.com/teortaxesTex/status/2108801260577284231", "who": "Teortaxes", "person_id": "teortaxes", "published_at": "10月10日", "evidence_role": "unknown", "event_id": null, "item_id": "41427", "source_role": "", "source_tier": ""}, {"id": "src:85989eab4e3d04afe401b4e4", "url": "https://x.com/MaxForAI/status/2102863004790579619", "who": "Max For AI", "person_id": "maxforai", "published_at": "9月24日", "evidence_role": "unknown", "event_id": "ev_plunging_price_of_thought", "item_id": "29440", "source_role": "", "source_tier": ""}, {"id": "src:7232eae48574fa341c6fa7be", "url": "https://x.com/teortaxesTex/status/2110829570052051256", "who": "Teortaxes", "person_id": "teortaxes", "published_at": "10月16日", "evidence_role": "unknown", "event_id": "ev_tower_intel_lawsuit", "item_id": "46349", "source_role": "", "source_tier": ""}, {"id": "src:f2bf0214344d0400cd97526f", "url": "https://x.com/tphuang/status/2110831297207865435", "who": "tphuang", "person_id": "tp_huang", "published_at": "10月16日", "evidence_role": "unknown", "event_id": "ev_tower_intel_lawsuit", "item_id": "46364", "source_role": "", "source_tier": ""}, {"id": "src:3fa164d071059062e29dacaf", "url": "https://x.com/MaxForAI/status/2110936912521560180", "who": "Max For AI", "person_id": "maxforai", "published_at": "2026-10-16T03:32:33.000Z", "evidence_role": "unknown", "event_id": null, "item_id": null, "source_role": "", "source_tier": ""}, {"id": "src:a1c7bf4f93d13cb93bd0f44e", "url": "https://x.com/tphuang/status/2110831297207865435", "who": "tphuang", "person_id": "tp_huang", "published_at": "2026-10-15T20:32:53.000Z", "evidence_role": "unknown", "event_id": null, "item_id": null, "source_role": "", "source_tier": ""}, {"id": "src:da221b57ea0dd73a6389b4c4", "url": "https://x.com/teortaxesTex/status/2110829570052051256", "who": "Teortaxes", "person_id": "teortaxes", "published_at": "2026-10-15T20:26:01.000Z", "evidence_role": "unknown", "event_id": null, "item_id": null, "source_role": "", "source_tier": ""}, {"id": "src:5822c3679fd6be8a133a5fb2", "url": "https://x.com/PrimeIntellect/status/2110779658723840142", "who": "Prime Intellect", "person_id": "prime_intellect", "published_at": "2026-10-15T17:07:41.000Z", "evidence_role": "unknown", "event_id": null, "item_id": null, "source_role": "", "source_tier": ""}, {"id": "src:815d2124b15bbdbf69ce6449", "url": "https://x.com/neal_k_patel/status/2110377281570886097", "who": "Neal Patel", "person_id": "neal_k_patel", "published_at": "2026-10-14T14:28:47.000Z", "evidence_role": "unknown", "event_id": null, "item_id": null, "source_role": "", "source_tier": ""}, {"id": "src:d7992386c67832a8e0d9ed0b", "url": "https://x.com/jerryjliu0/status/2108613049448108093", "who": "Jerry Liu", "person_id": "jerry_liu", "published_at": "2026-10-09T17:38:21.000Z", "evidence_role": "unknown", "event_id": null, "item_id": null, "source_role": "", "source_tier": ""}], "meta": {"built_at": "2026-10-19T16:35:31.958490+00:00", "material_cutoff": "2026-10-16T03:32:33.000Z", "tier": "anonymous", "window_days": 30, "windows": {"facts": {"start": "2026-09-10", "end": "2026-09-10"}, "claims": {"start": "2026-10-09T17:38:21.000Z", "end": "2026-10-16T03:32:33.000Z"}, "timeline": null}, "status": "partial", "included_counts": {"facts": 1, "issues": 3, "claims": 4, "relations": 0, "timeline": 0, "sources": 12, "who": 6}, "available_counts": {"facts": 1, "issues": 3, "claims": 4, "relations": 0, "timeline": null, "sources": 12, "who": 6}, "truncated": false, "omitted_reasons": ["permission_limited"], "continuation": null, "fallback_links": [], "errors": [], "timeline_hint": 36, "omitted_counts": {"facts": 0, "issues": 0, "claims": 0, "relations": 0, "timeline": null, "sources": 0, "who": 0}}, "slug": "inference_cost", "title": "推理成本", "kind": "tech", "one_liner": "推理 $/百万 token 的成本曲线，agent 放量的前提", "updated_at": "2026-09-10", "source": "Hourahead /o/inference_cost"}