Skip to content

A public ledger of token prices

Everything you spend on tokens disappears.So pick the cheapest hole.

Every plan is ranked by one number: how many dollars of usage, at DeepSeek's list price, each dollar you pay buys. Promotions, subscriptions, cache-hit rates and the fine print are all counted.

Get it in your inbox when the ledger moves

Buttondown sends a confirmation mail first. Unsubscribe from any mail.

The ledger

What one dollar actually buys.

Methodology →

16 of 16

Measured against
  1. 01

    Volcano Engine

    CheapestEstimated

    The cheapest place to run Kimi K3 right now

    volcengine.com·verified 2026-09-20

    Kimi K3, Doubao and other hosted models

    ¥49.00 → ¥800.00

    DeepSeek

    ~16.3x

    Mimo Pro

    —

    Mimo Flash

    —

  2. 02

    Alibaba Cloud

    Subscription

    TokenPlan Pro: $68 a month, double it off-peak

    alibabacloud.com·verified 2026-09-20

    Qwen and other hosted models

    $68.00 → $857.00

    DeepSeek

    12.6x

    Mimo Pro

    —

    Mimo Flash

    —

  3. 10x on selected models — but the context stops at 262k

    inferx.net·verified 2026-09-20

    200+ models; most endpoints capped at 262,000 context

    $20.00 → $200.00

    DeepSeek

    10x

    Mimo Pro

    —

    Mimo Flash

    —

  4. 04

    Command Code

    Subscription

    Go plan: the cheapest way in, at 10x

    commandcode.ai·verified 2026-09-20

    DeepSeek V4.1 Flash, GLM, Qwen and more

    $1.00 → $10.00

    DeepSeek

    10x

    Mimo Pro

    10x

    Mimo Flash

    10x

  5. 05

    Token Harbor

    Subscription

    Agent Pass: $0.99 first month, $10 of usage

    tokenharbor.ai·verified 2026-09-20

    GPT, Claude, Gemini, DeepSeek and more

    $0.99 → $10.00

    DeepSeek

    10x

    Mimo Pro

    10x

    Mimo Flash

    —

  6. 06

    Command Code

    Subscription

    Pro plan: $20 for $150 of credits

    commandcode.ai·verified 2026-09-20

    DeepSeek V4.1 Flash, GLM, Qwen, MiMo V2.6 and more

    $20.00 → $150.00

    DeepSeek

    7.5x

    Mimo Pro

    —

    Mimo Flash

    3.8x

  7. 07

    OpenDesign

    Subscription

    Go plan: $8 in the first month for $60 of usage — 7.5x

    open-design.ai·verified 2026-09-30

    DeepSeek V4.1 Flash, MiMo V2.6 Pro and Flash, GLM 5.3 Flash, Kimi, Hunyuan

    $8.00 → $60.00

    DeepSeek

    7.5x

    Mimo Pro

    7.5x

    Mimo Flash

    7.5x

  8. 08

    Command Code

    Subscription

    GOAT plan: $10 a month for $70 of model credits

    commandcode.ai·verified 2026-09-20

    DeepSeek V4.1 Flash, GLM, Qwen, Kimi, GPT-5.6, 30+ models

    $10.00 → $70.00

    DeepSeek

    7x

    Mimo Pro

    2x

    Mimo Flash

    6.7x

  9. 6.7x if you top up with BNB, 5x with anything else

    b.ai·verified 2026-09-25

    DeepSeek V4.1 Flash, GLM 5.3 Flash, MiMo V2.6 Pro and Flash

    $10.00 → $66.70

    DeepSeek

    6.7x

    Mimo Pro

    4x

    Mimo Flash

    4x

  10. 10

    OpenCode

    Subscription

    Go plan: $10 for $60 of DeepSeek V4.1 Flash a month

    opencode.ai·verified 2026-09-27

    DeepSeek V4.1 Flash and other open models

    $10.00 → $60.00

    DeepSeek

    6x

    Mimo Pro

    1.5x

    Mimo Flash

    6x

  11. 11

    Token Harbor

    Subscription

    Office Pass: $9.99 for $35 of usage

    tokenharbor.ai·verified 2026-09-20

    GPT, Claude, Gemini, DeepSeek and more

    $9.99 → $35.00

    DeepSeek

    3.5x

    Mimo Pro

    3.5x

    Mimo Flash

    —

  12. 3x today, 2x shortly — a rate that is on its way down

    llmgateway.io·verified 2026-09-20

    Routed across many providers

    $10.00 → $30.00

    DeepSeek

    3x

    Mimo Pro

    —

    Mimo Flash

    —

  13. 13

    Xiaomi

    Subscription

    Token Plan: do not buy it, the API is twice the value

    mimo.mi.com·verified 2026-09-30

    MiMo V2.6 Pro and V2.6 Flash (plus ASR and TTS)

    $6.00 → $5.95

    DeepSeek

    1.35x

    Mimo Pro

    —

    Mimo Flash

    —

  14. Pay as you go: MiMo V2.6 at half price, DeepSeek only 20% off

    openmodel.ai·verified 2026-09-28

    DeepSeek V4.1 Flash, MiMo V2.6 Pro/Flash, Claude, GPT, GLM, Kimi, Qwen and more

    $1.00 → $1.25

    DeepSeek

    1.25x

    Mimo Pro

    2x

    Mimo Flash

    2x

  15. Official pay-as-you-go — the baseline every deal is measured against

    api.deepseek.com·verified 2026-09-20

    DeepSeek V4.1 Flash, DeepSeek V4 Pro

    $10.00 → $10.00

    DeepSeek

    1x

    Mimo Pro

    —

    Mimo Flash

    —

  16. 16

    Hyper

    SubscriptionEstimated

    Off the ranking — billing moved to weekly

    hyper.charm.land·verified 2026-10-18

    Kimi K3, GLM 5.3 and GLM 5.3 flash

    $20.00 → $360.00

    DeepSeek

    ~18x

    Mimo Pro

    —

    Mimo Flash

    —

20x is not always better than 7x. Expiry and rate caps decide the real answer, so read the caveats.

Rank changes

What moved on the ledger

Every time a multiplier or a rank changed, with the reason. Ranks are on the DeepSeek column at the time.

  1. 2026-10-18

    HyperTaken down

    ~18xoff the board

    Off the board. Billing moved from monthly to weekly, and Hyper's app now says credits refresh every week at Sunday 00:00 UTC. The 18x was a monthly daily-refresh subscription, so it no longer describes what you pay. Volcano Engine takes first until this is re-rated.

  2. 2026-09-30

    B.AIRe-checked

    6.7x#10 → #10

    Filled in the MiMo columns: 4x on MiMo V2.6 Pro and Flash, but only when the top-up is paid in crypto.

  3. 2026-09-30

    7.5xnew at #8

    New: OpenDesign's Go plan, $8 for the first month and $60 of credits — 7.5x on DeepSeek, MiMo Pro and MiMo Flash alike.

  4. 2026-09-30

    1.35xnew at #13

    New: MiMo's Token Plan. Its own API at half price is 2x, so the subscription is not worth buying.

  5. 2026-09-28

    OpenModelChanged

    8x → 1.25x#7 ↓ #13

    DeepSeek V4.1 Flash is only 20% off ($0.12 / $0.48 vs the official $0.15 / $0.60), so 1.25x, not ~8x. MiMo V2.6 stays 2x at half price.

Full change log →

What to buy

What I would buy at each budget

One setup per budget, each step building on the last. This is my own call, not a ranking of multipliers.

Updated 2026-10-20

  1. $0 Free

    01

    Install CodeBuddy and use DeepSeek for free. At zero dollars, nothing else comes close.

    • CodeBuddyFree

      Which models are free changes with Tencent's billing — check the model list after you install.

  2. $1/mo

    02

    The cheapest way in: one dollar buys ten dollars of usage.

    • Command Code Go$1

      This plan is built for their own CLI. Wired into another tool as an API it draws a penalty, so use it through the CLI.

  3. $5/mo

    03

    The same Go plan on five accounts.

    • Command Code Go×5$5

      One plan per account. In my experience five accounts is still safe; past that, I would not risk it.

  4. $15/mo

    04

    Add ChatGPT Plus through GamsGo, far below the official $20.

    Total$15

  5. $35/mo

    05

    Claude Code on top. This is the setup I would actually run.

    Total$35

  6. Over $35

    Past this point it depends on how much you run. Compare multipliers on the ledger and build your own mix.

    Go to the ledger ↑

Free credits

Credit you can claim for signing up. Most of it is free; Hive's turns $5 into $100.

  • Modal

    $30/mo

    $30 of GPU compute every month on the free Starter plan. Run an open model yourself (vLLM and friends) and the tokens cost you nothing.

    It is compute, not tokens: you pay per GPU-second for what you host. Unused credit does not roll over.

    Claim it ↗
  • Hive

    $100 once

    $5 on the Hive Portal turns into $100 of credit once a payment method is on file. The API is OpenAI-compatible: DeepSeek V4.1 Flash and GLM 5.3 Flash are both on it.

    How many tokens that is

    • GLM 5.3 Flash2B60x
    • DeepSeek V4.1 Flash833M50x

    Input tokens at Hive's rate, and the multiplier against the list price it strikes through. Output costs 3-4x more, so a real mix lands below the token ceiling.

    What I saw in the portal; Hive's own pricing page only promises "$50+ in free credits after adding a payment method". A one-time sign-up promo.

    Claim it ↗

Model tiers

Coding model tiers

Ranked for coding only, from running them myself — not a benchmark, and price is not part of it. Within a tier, further left is better.

Updated 2026-10-20

Models marked ▶ have a recording of them building Super Mario — tap to watch.

  1. S

    1. Claude Opus 5.5 builds Super Mario

      Claude Code · medium

    2. Claude Fable 5.1 builds Super Mario

      high

  2. A

    1. Gemini 4 Argon builds Super Mario

      high · mario

    2. 2GPT-6 AstraOpenAI
    3. 3Claude Sonnet 5.5Anthropic
  3. A-

    1. 1GPT-6.1 SolOpenAI
    2. Kimi K3 builds Super Mario

  4. B+

    1. 1MiMo V2.6 ProXiaomi
    2. DeepSeek V4.1 Flash 1017 builds Super Mario

      max · mario

    3. GLM 5.3 builds Super Mario

      Z Code · max

    4. Grok 4.7 builds Super Mario

      bot · mario

    5. Claude Haiku 5.5 builds Super Mario

      xhigh · mario

  5. B

    1. Muse Wonder builds Super Mario

      high

    2. 2DeepSeek V4.1 FlashDeepSeek
    3. GLM 5.3 Flash builds Super Mario

      GLM 5.3 Flash

    4. 4Step 5 PreviewStepFun
  6. C+

    1. 1MiMo V2.6 FlashXiaomi
    2. 2Gemini 3.8 FlashGoogle
    3. MiniMax M3.1 Flash builds Super Mario

      Space Bunny · OpenCode max

  7. C

    1. 1Muse Spark 1.3Meta
    2. Mistral Large 4 Preview builds Super Mario

      high · mario

    3. 3MiniMax M3MiniMax
    4. GPT-6 Luna builds Super Mario

      max

  8. D+

    1. Muse Spark Open builds Super Mario

      high

  9. D

    1. Solar 4 ProUpstage
  10. D-

    1. MotiffKanyun

Best-value combos

The model mixes I think give the most for the money right now.

  1. No. 1

    • SClaude Opus 5.5
    • B+DeepSeek V4.1 Flash 1017
  2. No. 2

    • SClaude Opus 5.5
    • AGemini 4 Argon

Site reviews

What I actually think of each one.

The ledger gives you the number. These give you the verdict.

Read this before you top up

The fine print is the product.

The multiplier is the headline. Your bill is decided by expiry, rate caps, per-model allowances and whether the provider is still there next month.

Ask on the board

About this site

This site looks like AI slop because an AI built it. Every plan on it is one I actually pay for and use.

@aimaster3658on Threads