Q: How do the top 5 Chinese LLMs compare on price and performance in Q3 2026?

Top 5 Chinese LLMs ranked by per-million-token list price (CNY, as of 9/15): (1) DeepSeek V4.1 Flash — ¥1/M valley / ¥2/M peak (peak/valley pricing in effect since 9/10). Best price/performance at 128K context, 200K supported. (2) GLM-5.3-Flash — ¥0.8/M input / ¥2.8/M output. Zhipu flagship, strongest Chinese-language generation. (3) Qwen3.8-Flash — ¥0.4/M input / ¥1.2/M output. Alibaba cheapest, 1M context window, multimodal. (4) MiniMax-M3 — ¥1/M via tokenrhythm (¥1/¥1 USD price on tokenrhythm is the lowest for English). Best price/quality for English creative writing and tool-use. (5) Kimi K2.7 — ¥1/M input / ¥3/M output. Moonshot 256K context, agent-ready. 1M-context tier separately: DeepSeek V4-Pro-0813 / GLM-5.2 / qwen-3.8-max all support 1M tokens. For batch workloads: Qwen3.8-Flash wins on cost per million tokens (¥0.4 input). For code generation: DeepSeek V4.1 Flash leads on coding benchmarks. For Chinese long-form writing: GLM-5.3-Flash leads. For agent/research: Kimi K2.7 leads at 256K context. For multimodal (text+image+audio): Qwen3.8-Flash + MiniMax-M3 are best. Aggregators (tokenrhythm) can mix multiple models for ¥500 signup credit.

Q: What are the cheapest AI APIs in week 37 of 2026?

W37 (9/14-9/20) ranking Top 5 cheapest APIs by per-token list price: (1) **DeepSeek V4-Flash** ¥1/¥2 per 1M input/output — 1M context, cheapest in market. (2) **GLM-5.3-Flash** ¥0.4/¥1.4 — 1M context, Zhipu new-user 5-折 promo brings input to ¥0.2. (3) **Qwen3.8-Flash** ¥0.4/¥1.2 — 1M context, strongest Chinese-language. (4) **MiniMax-M3** ¥0.6/¥1.4 — best price/quality for English creative writing. (5) **Kimi K2.7** ¥1/¥3 — best for long-context reasoning with extended thinking. All five stack via tokenrhythm aggregator with 30-67% off list. 50M mixed tokens ≈ $14/month.

Q: What are the best AI subscription combinations in September 2026?

Five stacks under $80/month covering different workflows: (1) **Casual Chat** — ChatGPT Plus $20 + Perplexity Pro $20 = $40/month (covers GPT-5.6 Sol/Luna + real-time web search). (2) **Hardcore Coding** — Claude Max 5x ($100/yr÷12≈$8.3 annual) + GitHub Copilot Pro $10 + Devin Core $22 = $40.3/month (best for serious SWE work). (3) **Domestic Productivity** — Alibaba Bailian Qwen3.8-Flash ¥0.4/M + Volcano Doubao-2.0-pro ¥100 credit + DeepSeek V4.1 Flash valley ¥1/M via tokenrhythm = $68/month for 50M tokens. (4) **Full Multimodal** — MiniMax Speech 2.8 HD + Doubao Seedream 4.0 + Sora 2 subscription = $70/month. (5) **Research & Agent** — Perplexity Max $200 + Devin Core $22 = $222/month, or budget alternative tokenrhythm ¥500 credit + Kimi K3 Agent API = $30/month. Annual billing adds 17% savings on Cursor Pro, ChatGPT Plus, Claude Pro.

Question 1

本周 AI 厂商最大价格变动（2026 年 9 月第 3 周）：哪些值得抄底？

Accepted Answer

Biggest vendor pricing moves week 38 (Sep 13-19, 2026): (1) **DeepSeek V4.1 Flash peak-valley extended 7 days to 9/24** — valley ¥1/M input, peak ¥3/M input. For a 50M token/month agent workload with 60% peak ratio, switching to 100% valley saves ¥1500/month. (2) **GLM-5.3-Flash 5-折 extended to 9/30** — ¥0.4/¥1.4 vs V4-Flash ¥1/¥2 saves ¥30/M input. Cheapest tier for chat / writing. (3) **MiniMax-M3 joins TokenRhythm 0.5-折 first month** — ¥0.5/¥0.5 input/output (vs ¥1/¥1 official). Limited to first 1,000 new users. (4) **Volcano doubao-2.0-pro ¥100 credit new user signup** — covers 100M input tokens. (5) **iFlytek Spark V4.5 5-折 ¥0.7/¥2.1 input** — second cheapest tier after GLM/Qwen Flash. (6) **OpenAI / Anthropic / Google flagship prices stable** — GPT-5.6 $2.50/$10, Claude Sonnet 5 $3/$15, Gemini 3.7 Flash $0.075/$0.30. Net arbitrage for an agent-heavy user running 100M tokens/month: ¥2400/mo savings on a ¥4000 baseline. Recommended picks: **coding** → Cursor Pro $20; **writing** → Claude Sonnet 5; **batch workloads** → DeepSeek V4.1 Flash valley; **multimodal** → Gemini 3.7 Flash; **research / 1M context** → GLM-5.2 1M; **chat** → GLM-5.3-Flash 5-折. Stack with TokenRhythm ¥500 free credit (extended to 9/30) for first-month $0 effective cost on multi-model workloads.

Question 2

Coding Plan vs Token Plan vs API 聚合平台：2026 年 9 月该怎么选？

Accepted Answer

Three subscription models coexist in Q3 2026, each fitting different workflows: (1) **Coding Plans** — IDE-integrated fixed-price like Cursor Pro $20/mo, GitHub Copilot Pro $10/mo, GLM Coding Plan ¥118-1078/mo, MiniMax Coding Plus ¥49/mo, Kimi Code ¥49-699/mo, Claude Code Max $100-200/mo. Best for individual developers who want one-click IDE experience. (2) **Token Plans** — metered subscriptions like Alibaba Bailian Lite ¥39/mo, Tencent HY3 ¥28/mo, Volcano Engine BYO model. Best for production workloads with predictable token spend; gives bulk discount on pay-as-you-go. (3) **API Aggregators** — multi-vendor single key like TokenRhythm (¥500 Sep credit + ¥68 signup), OpenRouter (50+ models), SiliconFlow. Best for cost optimization, regional arbitrage, and fail-over. Pick **Coding Plan** for daily dev convenience, **Token Plan** for cost control at scale (>30M output tokens/mo), **Aggregator** for price arbitrage (<50M tokens/mo or one-off testing). Wednesday combo recommendation: Claude Max 5x + GitHub Copilot Pro + Devin Core = $40.3/mo for hardcore coding + TokenRhythm ¥500 credit for cheap API tests.

Question 3

How do the top 5 Chinese LLMs compare on price and performance in Q3 2026?

Accepted Answer

Top 5 Chinese LLMs ranked by per-million-token list price (CNY, as of 9/15): (1) DeepSeek V4.1 Flash — ¥1/M valley / ¥2/M peak (peak/valley pricing in effect since 9/10). Best price/performance at 128K context, 200K supported. (2) GLM-5.3-Flash — ¥0.8/M input / ¥2.8/M output. Zhipu flagship, strongest Chinese-language generation. (3) Qwen3.8-Flash — ¥0.4/M input / ¥1.2/M output. Alibaba cheapest, 1M context window, multimodal. (4) MiniMax-M3 — ¥1/M via tokenrhythm (¥1/¥1 USD price on tokenrhythm is the lowest for English). Best price/quality for English creative writing and tool-use. (5) Kimi K2.7 — ¥1/M input / ¥3/M output. Moonshot 256K context, agent-ready. 1M-context tier separately: DeepSeek V4-Pro-0813 / GLM-5.2 / qwen-3.8-max all support 1M tokens. For batch workloads: Qwen3.8-Flash wins on cost per million tokens (¥0.4 input). For code generation: DeepSeek V4.1 Flash leads on coding benchmarks. For Chinese long-form writing: GLM-5.3-Flash leads. For agent/research: Kimi K2.7 leads at 256K context. For multimodal (text+image+audio): Qwen3.8-Flash + MiniMax-M3 are best. Aggregators (tokenrhythm) can mix multiple models for ¥500 signup credit.

Question 4

What are the cheapest AI APIs in week 37 of 2026?

Accepted Answer

W37 (9/14-9/20) ranking Top 5 cheapest APIs by per-token list price: (1) **DeepSeek V4-Flash** ¥1/¥2 per 1M input/output — 1M context, cheapest in market. (2) **GLM-5.3-Flash** ¥0.4/¥1.4 — 1M context, Zhipu new-user 5-折 promo brings input to ¥0.2. (3) **Qwen3.8-Flash** ¥0.4/¥1.2 — 1M context, strongest Chinese-language. (4) **MiniMax-M3** ¥0.6/¥1.4 — best price/quality for English creative writing. (5) **Kimi K2.7** ¥1/¥3 — best for long-context reasoning with extended thinking. All five stack via tokenrhythm aggregator with 30-67% off list. 50M mixed tokens ≈ $14/month.

Question 5

What are the best AI subscription combinations in September 2026?

Accepted Answer

Five stacks under $80/month covering different workflows: (1) **Casual Chat** — ChatGPT Plus $20 + Perplexity Pro $20 = $40/month (covers GPT-5.6 Sol/Luna + real-time web search). (2) **Hardcore Coding** — Claude Max 5x ($100/yr÷12≈$8.3 annual) + GitHub Copilot Pro $10 + Devin Core $22 = $40.3/month (best for serious SWE work). (3) **Domestic Productivity** — Alibaba Bailian Qwen3.8-Flash ¥0.4/M + Volcano Doubao-2.0-pro ¥100 credit + DeepSeek V4.1 Flash valley ¥1/M via tokenrhythm = $68/month for 50M tokens. (4) **Full Multimodal** — MiniMax Speech 2.8 HD + Doubao Seedream 4.0 + Sora 2 subscription = $70/month. (5) **Research & Agent** — Perplexity Max $200 + Devin Core $22 = $222/month, or budget alternative tokenrhythm ¥500 credit + Kimi K3 Agent API = $30/month. Annual billing adds 17% savings on Cursor Pro, ChatGPT Plus, Claude Pro.

AI 太贵？我们帮你省钱

快速推荐

热门模型当前价格

最新价格变动

Coding 方案

按场景选 AI

深度指南

成本计算器

优惠折扣

价格变动历史