🌍A 65.3FlagshipClosed
Anthropic
Anthropic于2026年9月22日发布的Opus档最新旗舰模型，在多数工作上达到Claude Fable 5.1的水平，而运行成本比Opus 5低约40%。主打长时程高自主性Agent、大规模代码库的构建/调试/重构/代码审查，以及端到端企业工作流；支持记忆跨会话学习、子Agent编排与Fast模式（最高2.5倍速）。
Composite score65.3/100
Capability100
Scale37
Value34
Context61
Ecosystem36
Freshness89
Capability LeaderNew ReleaseLong Context
Output / 1M tokens$20USD
Input $4
※ 官方产品页（anthropic.com/claude/opus，2026-10-24 复核）：输入 $4 / 缓存读取 $0.2 / 输出 $20 每百万 tokens，比 Opus 5 低 20%，缓存读取便宜 60%；Fast 模式 $8/$40；美国境内推理按 1.1 倍计价；Batch 五折
vs_Claude_Fable_5.1 多数工作负载持平vs_Opus_5_cost 典型工作负载成本 -40%
大规模代码库Agent编程端到端企业工作流长时程自主Agent
How to access
Claude API（模型名 claude-opus-5-5）/ AWS Bedrock / Google Cloud / Microsoft Foundry；含于 Pro、Max、Team、Enterprise 订阅。中国大陆无法直接使用。
智谱AI 2026年8月14日发布的新一代旗舰模型，743B参数，与GLM-5.2同基座，通过后训练Scaling（IndexShare、SAO及开源Slime强化学习框架）编程能力较GLM-5.2提升50%，Terminal Bench 3.0、Agents' Last Exam (CLI) 开源第一，Agent能力与Kimi K3、Claude Opus 4.8、Claude Fable 5同一水平。已上线ZCode、AutoClaw及GLM Coding Plan全量用户，完整权重两周内开源，API即将上线。
7×cheaper than GPT-5.5 (output $/M)
Composite score62.1/100
Capability79
Scale29
Value64
Context41
Ecosystem70
Freshness68
Output / 1M tokens$4.4USD
Input $1.4
※ 海外定价（Z.ai 官方定价页 https://docs.z.ai/guides/overview/pricing ，2026-10-24 复核）：$1.4 输入 / $4.4 输出 每百万 tokens（缓存命中 $0.26，缓存存储限时免费）。
Terminal_Bench_3.0 Open-source #1Agents_Last_Exam_CLI Open-source #1
AI编程（ZCode/Claude Code等编码平台）长周期智能体任务复杂软件工程
How to access
已上线 ZCode / AutoClaw / GLM Coding Plan；权重已开源，API 经 open.bigmodel.cn 与 z.ai 提供。
月之暗面2026年7月发布的2.8T参数前沿旗舰模型，896专家MoE架构、每token激活16个专家，1M上下文统一定价无分级加价。KDA混合注意力机制带来6.3倍解码加速，前端编程能力全球第一，编码与推理对标Claude Opus而成本低一半。2026年7月27日开放权重。
2×cheaper than GPT-5.5 (output $/M)
Parameters
2.8T MoE (16 experts active)
Composite score56.7/100
Capability73
Scale35
Value37
Context61
Ecosystem63
Freshness58
Long Context
Output / 1M tokens$15USD
Input $3
※ Overseas pricing (Moonshot/Kimi official pricing page): $3.00 input (cached input $0.30) / $15.00 output per 1M tokens. Mainland-China pricing (¥20/¥100) is shown on the Chinese page
Frontend_coding #1 globallyArtificial_Analysis_Index Top tier
编码Agent长上下文推理复杂多步骤任务
How to access
API 见 platform.moonshot.cn 与 kimi.ai。兼容 OpenAI SDK——迁移成本几乎为零。开放权重计划 7 月 27 日发布。
🌍B 54.3FlagshipClosed
OpenAI
OpenAI于2026年7月9日发布的新一代模型家族，分为三个层级：旗舰级Sol、主力型号Terra和轻量级Luna。Luna旨在实现低成本高吞吐量，其事实性错误比GPT-5.5 Instant少约62%。自7月30日起永久降价：Luna降价80%，Terra降价20%；Sol新增Fast模式（价格翻倍，速度提升2.5倍）。
Composite score54.3/100
Capability43
Scale35
Value85
Context76
Ecosystem44
Freshness48
Long Context
Output / 1M tokens$1.2USD
Input $0.2
※ Luna pricing; permanent cuts from July 30: Terra $2/$12 (-20%), Luna $0.2/$1.2 (-80%); Sol Fast mode 2x price for 2.5x speed
Artificial Analysis Luna beats DeepSeek V4-Pro on cost-efficiency
批量文本处理高吞吐量智能体企业级智能体
How to access
可通过platform.openai.com直接API访问。中国大陆地区不可直接使用。
DeepSeek 旗舰级 MoE 模型（模型版本 DeepSeek-V4-Pro-0813），支持 1M 上下文与 384K 最大输出，默认开启思考模式，支持 JSON 输出、工具调用、Responses API 与 Anthropic API 兼容格式。
15×cheaper than GPT-5.5 (output $/M)
Parameters
1.6T MoE (49B active)
Composite score49.3/100
Capability40
Scale35
Value73
Context77
Ecosystem78
Freshness0
Long Context
Output / 1M tokensOff-peak$1.98USD
Input $0.66
⚡ Peak Input $1.32 · Output $3.9601:00–04:00 / 06:00–10:00 UTC, Mon–Fri
※ 海外定价（DeepSeek 官方国际定价页 https://api-docs.deepseek.com/quick_start/pricing/ ，2026-10-24 复核）：非高峰 $0.66 输入（缓存命中 $0.022）/ $1.98 输出 每百万 tokens，为标准高峰价（$1.32/$3.96）的一半。高峰时段为 UTC 周一至周五 01:00–04:00 与 06:00–10:00（不含中国法定节假日），其余时段含周末与节假日均为非高峰。。网页搜索：Search / Fetch API 各 ¥5 / 1,000 次成功请求（≈$0.74，次均 ¥0.005）；在 DeepSeek Harness / Claude Code / Codex 中经 Server 端工具调用 Search / Fetch 目前免费，自 2026-11-01 00:00（北京时间）起按次计费
SWE-bench_Verified 80.6%Codeforces 3206
复杂推理长文档分析代码生成
How to access
通过 api.deepseek.com 直接调用，支持 OpenAI 与 Anthropic 两种格式；需中国支付方式或国际信用卡充值。
🌍D 34.8FlagshipClosed
Google
Google's 2026 flagship model with 1M-token context and advanced multimodal capabilities. Released on July 15, 2026, focusing on enterprise applications and developer tools.
Composite score34.8/100
Capability31
Scale35
Value31
Context35
Ecosystem51
Freshness51
Output / 1M tokens$12USD
Input $2
※ Per Google Cloud Vertex official pricing (Gemini Pro tier, global): $2.00/$12.00, matching Artificial Analysis (gemini-3-pro, $2/$12). The previous $5/$15 was mis-entered during the 2026-09-23 data recovery with no source; corrected 2026-10-11
Enterprise applicationsDeveloper toolsMultimodal content processing
How to access
Direct API access via Google Cloud. Available in most regions including mainland China via international services.
🇨🇳D 27.9BudgetClosed
ByteDance
字节跳动2026年2月14日发布的旗舰大模型，IMO/CMO数学竞赛和ICPC编程竞赛金牌水平，超越Gemini 3 Pro的Putnam基准。多模态、科学推理、Agent全能，月活破亿。
13×cheaper than GPT-5.5 (output $/M)
Composite score27.9/100
Capability35
Scale35
Value35
Context30
Ecosystem51
Freshness0
Output / 1M tokens$2.35USD
Input $0.47
※ Domestic pricing (Volcano Engine Ark console model page, checked 2026-10-11): ¥3.2 input / ¥16 output per 1M tokens (≤32K input tier; tiered billing above that). No USD price exists — this model is China-only: the BytePlus international console returns 'the current region does not currently support accessing this page' for doubao-seed-2-0-pro / seed-2-0-pro, and OpenRouter carries only the Code/Lite/Mini siblings, not Pro. The English page therefore shows the domestic CNY price
IMO_CMO Gold medalICPC Gold medal
数学竞赛级推理多模态内容创作日常对话（月活破亿）
How to access
经字节跳动火山引擎提供。以中国为主；全球访问受限。
🌍FlagshipClosed
Google Dee
Google's newest frontier model, released October 14, 2026, built for the hardest real-world tasks in coding, finance, law, cybersecurity and science. It ranks #1 on the Vals Index for economically valuable work, supports up to 1,048,576 output tokens sharing a single 1M-token context window with input and thinking, and cuts harmful-request violation rates by nearly half versus the previous Gemini generation.
Max Output
1,048,576 Tokens
Output / 1M tokens$10USD
Input $2
※ Limited-time introductory pricing announced at launch (2026-10-14): $2 input / $10 output / $0.10 cached input per 1M tokens. After the introductory promotion ends on 2026-11-30, pricing rises to $4 input / $20 output / $0.20 cached input
Vals_Index #1 (finance, coding, legal, tax)
Large-scale codebase refactoringLegal and financial workflow automationCybersecurity vulnerability discovery and remediation
How to access
Available worldwide to developers and enterprises via Google AI Studio and the Gemini platform API; also included for Google AI Ultra subscribers on Google Antigravity and for AI Ultra/Pro subscribers in the Gemini app and Search AI Mode. No Chinese payment method required.
Meta's flagship frontier model released October 13, 2026, which Mark Zuckerberg says puts the company back at the AI frontier. Meta claims it outperforms Anthropic's Claude Opus 5.5 and Fable 5.1 and OpenAI's GPT-6 Astra on some technical benchmarks, with strong advanced coding and improved multi-turn memory. It is closed-source, with the strongest tier sold directly to enterprises.
Output / 1M tokens$0.2USD
Input $0.2
※ Meta announced tiered pricing at launch (2026-10-13): basic calls start at a minimum of $0.20 per 1M tokens, while the higher-capability Muse Wonder Full tier is charged up to $200 per month. Meta has not yet published a complete per-token input/output price list, so the input and output fields show the announced $0.20/1M-token floor
vs_Claude_Opus_5.5_Fable_5.1_GPT-6_Astra Outperforms on some benchmarks (Meta claim)
Advanced software engineeringLease contract and financial document analysisHealth and financial information synthesis
How to access
Closed-source; the strongest tier is sold directly to enterprises. Meta plans to integrate the model into its Muse personal AI assistant app. No Chinese payment method required.
Why global developers are switching to Chinese AI models
Since 2025, Chinese LLMs have surged from 4.5% to over 46% of US enterprise tokens on OpenRouter. The driver is simple economics: models like DeepSeek V4 and GLM-5.2 deliver near-frontier quality at 60–90% lower cost than GPT-5.5 or Claude Opus. Companies like Lindy, Coinbase, DoorDash and Siemens have already made the switch.
How to choose the right Chinese LLM
For coding agents and long-context reasoning, DeepSeek V4 Pro leads. For self-hosting and multilingual apps, Qwen3 Max offers the strongest open-weight ecosystem. GLM-5.2 is the cheapest Claude replacement, Kimi K2.5 owns the 2M-token long-context niche, and MiniMax M2.5 / Doubao 2.0 win on raw throughput cost. Filter by origin, open weights, or price above.
How to access Chinese AI models from outside China
Most Chinese frontier models are available globally without a Chinese payment method. DeepSeek, Qwen, GLM and MiniMax all accept international cards or run on OpenRouter and Together.ai. Qwen, DeepSeek and GLM also ship open weights on Hugging Face for full self-hosting. Each card above lists exact access steps.