AI Daily Insights No. 102 · November 14, 2026
←All topics

China model API pricing

Current prices checked against official pages; every change sourced

Where our earlier coverage reported conflicting figures, we left them out until verified. Qwen, Zhipu and MiniMax are listed at the CNY prices on their official China price pages; international USD prices differ.

DATA UPDATED 2026-11-02 · CHECK DATES PER ROW · CALCULATION GUIDE UPDATED 2026-10-24

API cost calculator

Enter monthly usage to see what each model costs per month at official prices (check dates are shown on each row below), in CNY, cheapest first. Models priced in USD are converted at 1 USD = 6.733 CNY (PBoC central parity, 2026-10-09).

  1. OpenAIgpt-6-luna *
    135 CNY / monthinput ¥67.33 · output ¥67.33
  2. AnthropicClaude Haiku 5.5 *
    135 CNY / monthinput ¥67.33 · output ¥67.33
  3. ZhipuGLM-5.3-Flash *
    136 CNY / monthinput ¥80 · output ¥56
  4. BaiduERNIE-4.5-Turbo *
    144 CNY / monthinput ¥80 · output ¥64
  5. MiniMaxMiniMax-M3.1-Flash
    180 CNY / monthinput ¥100 · output ¥80
  6. DeepSeekdeepseek-flash *
    270 CNY / monthinput ¥150 · output ¥120
  7. Qwenqwen3.7-plus *
    360 CNY / monthinput ¥200 · output ¥160
  8. MiniMaxMiniMax-M3 *
    378 CNY / monthinput ¥210 · output ¥168
  9. MiniMaxMiniMax-M2.7 *
    378 CNY / monthinput ¥210 · output ¥168
  10. GoogleGemini 3.5 Flash-Lite
    539 CNY / monthinput ¥202 · output ¥337
  11. Mistral AIMistral Large *
    539 CNY / monthinput ¥337 · output ¥202
  12. BaiduERNIE-5.1 *
    760 CNY / monthinput ¥400 · output ¥360
  13. GoogleGemini 3.8 Flash *
    1,010 CNY / monthinput ¥505 · output ¥505
  14. DeepSeekdeepseek-v4-pro *
    1,080 CNY / monthinput ¥675 · output ¥405
  15. xAIgrok-4.3 *
    1,178 CNY / monthinput ¥842 · output ¥337
  16. Kimikimi-k2.7-code *
    1,190 CNY / monthinput ¥650 · output ¥540
  17. Kimikimi-k2.6 *
    1,190 CNY / monthinput ¥650 · output ¥540
  18. ZhipuGLM-5.3
    1,360 CNY / monthinput ¥800 · output ¥560
  19. Mistral AIMistral Large 4 *
    1,479 CNY / monthinput ¥916 · output ¥563
  20. Qwenqwen3.8-max *
    1,920 CNY / monthinput ¥1,200 · output ¥720
  21. xAIgrok-4.7 *
    2,155 CNY / monthinput ¥1,347 · output ¥808
  22. Kimikimi-k3.1-pro *
    2,660 CNY / monthinput ¥1,300 · output ¥1,360
  23. OpenAIgpt-6.1-sol *
    2,693 CNY / monthinput ¥1,347 · output ¥1,347
  24. AnthropicClaude Sonnet 5.5
    2,693 CNY / monthinput ¥1,347 · output ¥1,347
  25. GoogleGemini 4 Argon *
    2,693 CNY / monthinput ¥1,347 · output ¥1,347
  26. GoogleGemini 3.1 Pro Preview *
    2,963 CNY / monthinput ¥1,347 · output ¥1,616
  27. Kimikimi-k3 *
    4,000 CNY / monthinput ¥2,000 · output ¥2,000
  28. Kimikimi-k3.1-pro-highspeed *
    5,320 CNY / monthinput ¥2,600 · output ¥2,720
  29. AnthropicClaude Opus 5.5
    5,386 CNY / monthinput ¥2,693 · output ¥2,693
  30. OpenAIgpt-6-astra *
    13,466 CNY / monthinput ¥6,733 · output ¥6,733
  31. AnthropicClaude Fable 5.1
    13,466 CNY / monthinput ¥6,733 · output ¥6,733

* Conditions apply, see table notes. The peak share only affects models with time-of-day pricing, such as DeepSeek. Tiered prices and limited-time discounts are in the notes of the table below; this uses the prices as listed there and excludes cache-write, storage and tool-call charges.

How to check an API cost calculation

Split input into cache misses and cache hits, then calculate output separately. The formula below uses raw token counts; the calculator above takes millions of tokens, so do not divide those inputs by a million again.

Cost = (uncached input tokens × input rate + cached input tokens × cache rate + output tokens × output rate) ÷ 1,000,000

Worked example: illustrative rates, not a provider quote

Assume rates in one currency of $1 per million uncached input tokens, $0.10 per million cached input tokens, and $4 per million output tokens. One request has 1,000 input tokens, including 200 cache hits, and generates 500 output tokens.

ComponentTokens × rate ÷ 1,000,000Cost
Uncached input800 × $1 ÷ 1,000,000$0.00080
Cached input200 × $0.10 ÷ 1,000,000$0.00002
Output500 × $4 ÷ 1,000,000$0.00200

Total: $0.00282 per request. At 10,000 requests a day for 30 days with the same usage per request, the total is $846. Request volume must be included too.

This is not a bill guarantee. Time-of-day pricing must be calculated from actual billable tokens in each period; the share of calls is only an estimation assumption. Check region, long-context tiers, discounts, how reasoning tokens are billed, and extra search, tool, cache-write or storage charges. No listed cache rate does not mean free caching. Compare only within the applicable billing conditions.

Verified rates (check date on each row)

CNY per million tokens. Where time-of-day pricing applies: peak / off-peak.

Full baseline checked 2026-10-14; later rechecks are dated per row. Older dates are retained. — means no separate cache price listed.
ModelInput · cache missInput · cache hitOutputSource
DeepSeekdeepseek-flash（V4.1-Flash）International USD: input $0.15 / $0.30, output $0.60 / $1.20 (off-peak / peak); peak = 01:00–04:00 and 06:00–10:00 UTC on weekdays, excluding Chinese public holidays 2 / 10.04 / 0.028 / 4Official ↗Checked 2026-10-24
DeepSeekdeepseek-v4-pro（V4-Pro-0813）International USD: input $0.66 / $1.32, output $1.98 / $3.96 (off-peak / peak) 9 / 4.50.3 / 0.1527 / 13.5Official ↗Checked 2026-10-24
Moonshot Kimikimi-k3.1-proStandard model. Cache writes: ¥13/M for 5-minute TTL or ¥26/M for 1 hour; estimates exclude cache-write charges. 131.368Official ↗Checked 2026-10-24
Moonshot Kimikimi-k3.1-pro-highspeedHighspeed model. Cache writes: ¥26/M for 5-minute TTL or ¥52/M for 1 hour; estimates exclude cache-write charges. 261.3136Official ↗Checked 2026-10-24
Moonshot Kimikimi-k3Cache writes have separate 5-minute / 1-hour rates, excluded from these cost estimates. 202100Official ↗Checked 2026-10-24
Moonshot Kimikimi-k2.7-codeK2 access for newly registered users ended October 31; existing registered users are unaffected. Retirement remains planned for late December. 6.51.327Official ↗Checked 2026-10-24
Moonshot Kimikimi-k2.6K2 access for newly registered users ended October 31; existing registered users are unaffected. Retirement remains planned for late December. 6.51.127Official ↗Checked 2026-10-24
Alibaba Qwenqwen3.8-maxinput ≤1M tokens, Beijing region 12—36Official ↗Checked 2026-10-24
Alibaba Qwenqwen3.7-pluslist price for input ≤256K; currently 20% off for a limited time 2—8Official ↗Checked 2026-10-24
ZhipuGLM-5.3 8228Official ↗Checked 2026-10-24
ZhipuGLM-5.3-Flashstandard price; currently 50% off for a limited time 0.80.232.8Official ↗Checked 2026-10-24
MiniMaxMiniMax-M3.1-Flash 10.14Official ↗Checked 2026-10-24
MiniMaxMiniMax-M3input ≤512K, after the official permanent 50% discount 2.10.428.4Official ↗Checked 2026-10-24
MiniMaxMiniMax-M2.7Rates last verified October 14. The official overview still did not list this model on November 2; current availability and rates are unconfirmed. Recheck before use. 2.10.428.4Official ↗Checked 2026-10-14
百度 文心ERNIE-5.1input ≤32K; 32K–128K is ¥6 input / ¥22 output 4—18Official ↗Checked 2026-10-24
百度 文心ERNIE-4.5-Turbosame price for 32K and 128K; batch: ¥0.32 input / ¥1.28 output 0.80.23.2Official ↗Checked 2026-10-24

International models (official pages, checked 2026-10-24)

USD per million tokens. Standard rates for the models covered here; long-context and other conditions are in the notes.

ModelInputInput · cache hitOutputSource
OpenAIgpt-6-astraStandard tier, prompts up to 272K tokens 10150Official ↗Checked 2026-10-24
OpenAIgpt-6.1-solStandard tier, prompts up to 272K tokens 20.110Official ↗Checked 2026-10-24
OpenAIgpt-6-lunaStandard tier, prompts up to 272K tokens 0.10.010.5Official ↗Checked 2026-10-24
AnthropicClaude Fable 5.1 100.2550Official ↗Checked 2026-10-24
AnthropicClaude Opus 5.5 40.220Official ↗Checked 2026-10-24
AnthropicClaude Sonnet 5.5 20.110Official ↗Checked 2026-10-24
AnthropicClaude Haiku 5.5Prompts up to 100K tokens; above that $0.50 input, $0.05 cache hit, $2.50 output 0.10.010.5Official ↗Checked 2026-10-24
GoogleGemini 4 ArgonStandard introductory rates through November 30, 2026; from December 1 input $4, cached input $0.20, output $20. Implicit caching only; tool calls billed separately. 20.110Official ↗Checked 2026-10-24
GoogleGemini 3.1 Pro PreviewPrompts up to 200K tokens; above that $4 input, $0.40 cached, $18 output 20.212Official ↗Checked 2026-10-24
GoogleGemini 3.8 FlashThrough 2026-12-31; from 2027-01-01 $1.50 input, $0.15 cached, $7.50 output 0.750.0753.75Official ↗Checked 2026-10-24
GoogleGemini 3.5 Flash-Lite 0.30.032.5Official ↗Checked 2026-10-24
xAIgrok-4.7Prompts under 200K tokens; at 200K+ the whole request bills at $4 input, $1 cached, $12 output 20.56Official ↗Checked 2026-10-24
xAIgrok-4.3Prompts under 200K tokens; prices double at 200K+ 1.250.22.5Official ↗Checked 2026-10-24
MistralMistral Large 4Public preview. Official Standard rates; other service tiers have separate rates. 1.360.144.18Official ↗Checked 2026-10-24
MistralMistral LargeThis row is Mistral Large 3; batch is 50% off and cached input is $0.05/M tokens. 0.50.051.5Official ↗Checked 2026-10-24

Vendor by vendor

Price change log

  1. 2026-10-24

    Verification date, not a shared effective date: Kimi lists K3.1 Pro Standard / Highspeed at ¥13 / ¥26 input and ¥68 / ¥136 output, with ¥1.3 cache hits per million tokens. K2 access restrictions and retirement remain future plans. Source · Our issue

  2. 2026-10-24

    Verification date: MiniMax lists M3.1-Flash at ¥1 input / ¥0.1 cached input / ¥4 output per million tokens; existing M3 standard rates are unchanged. This is not an asserted launch or repricing date. Source

  3. 2026-10-24

    Verification date: Mistral Large 4 Standard is $1.36 input / $0.14 cached input / $4.18 output. The existing Large row is Large 3; its listed $0.05 cache-read rate has been added. Source

  4. 2026-10-24

    Verification date: Gemini 4 Argon Standard introductory rates are $2 input / $0.10 cached input / $10 output through November 30. Listed rates from December 1 are $4 / $0.20 / $20, not yet the current rates. Source

  5. 2026-10-24

    Verification date: DeepSeek model token rates are unchanged. Search and Fetch are separately listed at ¥5 per 1,000 successful requests each; server-side tool calls in Harness and similar clients start per-request billing at 00:00 Beijing time on November 1. These tool charges are outside this token calculator. Source

  6. 2026-04-24

    DeepSeek V4-Pro launched at a list price of ¥12 input (cache miss) / ¥24 output, with a 75% launch discount (¥3 / ¥6). Source

  7. 2026-04-26

    DeepSeek cut cache-hit input prices across the lineup to a tenth of launch levels: ¥0.02 for V4-Flash, ¥0.025 for V4-Pro. Source

  8. 2026-05-31

    DeepSeek said V4-Pro would stay permanently at a quarter of its list price after the promotion — ¥3 input / ¥6 output (about $0.435 / $0.87). Source

  9. 2026-07-16

    Kimi K3 launched (≈2.8T parameters, 1M-token context). International list price: $3 input / $15 output per million tokens (our Aug 27 issue). The China site now lists ¥20 input / ¥100 output. Source · Our issue

  10. 2026-08-17

    DeepSeek introduced peak/off-peak API pricing: weekday 9:00–12:00 and 14:00–18:00 Beijing time are peak; off-peak is half price. V4 Pro output ¥27 peak / ¥13.5 off-peak; V4 Flash output ¥9 / ¥4.5. Source · Our issue

  11. 2026-08-16

    International (USD) peak/off-peak pricing took effect: V4 Flash input $0.22 / $0.44, output $0.66 / $1.32; V4 Pro input $0.66 / $1.32, output $1.98 / $3.96 (off-peak / peak). Source

  12. 2026-09-10

    DeepSeek launched V4.1 Flash and said it planned to route V4 Pro requests to the new model at Flash prices. Source · Our issue

  13. 2026-09-30

    Official price page check: deepseek-flash (V4.1-Flash) output is ¥8 peak / ¥4 off-peak ($1.20 / $0.60 internationally), below V4 Flash’s ¥9 / ¥4.5 ($1.32 / $0.66) in August; no announcement of the change date was found. v4-pro is unchanged since Aug 17. Source

  14. 2026-09-30

    Kimi K3.1 appeared on the API platform; the official price page does not list it yet. Source · Our issue