AICost

Bảng giá API LLM độc lập: so sánh giá mô hình giữa OpenAI, Anthropic, Google và nhiều nhà cung cấp khác.

Tài liệu

The independent price desk for every AI model on the market

The same 82 work tasks cost $0.079 on one model and $4.09 on another.

AI Cost Bench is our own test. 10 models run identical tasks through their makers' APIs, a program checks every answer, and we add up what each was billed: 52x between the cheapest run and the most expensive. The index counts tasks done right per dollar, with 10 cents charged for every task a model got wrong: Muse Ember leads at 271.

AI Cost Bench Index tasks done right per dollar

  1. 271 Muse Ember (max)
  2. 199 Claude Haiku 5.5 (medium)
  3. 129 GPT-6.1 Terra (medium)
  4. 102 GLM-5.3-FlashX (max)
  5. 90 MiniMax M3.1 Flash (max)
  6. 76 GPT-6 Luna (medium)
  7. 68 Qwen3.7 Plus
  8. 47 Claude Sonnet 5.5 (high)
  9. 20 Gemini 4 Hydrogen (medium)
  10. 18 Gemini 3.5 Flash-Lite (minimal)

Higher is better · in brackets: default reasoning effort

§ 02 / INDEX

The full desk.

Sitemap →

SectionCoveragePagesUpdatedStatus
Pricing Hub →Per-model pricing with calculators and historical charts354 pagesNov 20LIVE
Benchmarks →Leaderboards for the tests labs rank models on, every score linked to its source, with live API prices35 leaderboardsNov 8NEW
AI Subscriptions →Chat and coding plans side by side, every price read on the vendor's own page, with what each tier includes13 products / 88 plansNov 18NEW
Comparisons →Two-way head-to-heads with live pricing, benchmarks, and verdicts56 pagesNov 18LIVE
Capability Q&A →"Can X do Y?" with verified test outputs and screenshots61 pagesOct 3LIVE
Calculators →LLM API cost, token counting, subscription vs API, video generation, text-to-speech, embeddings, fine-tuning, agent cost and image generation — on live rates9 live toolsNov 20NEW
AI Agents →Every route is priced from the vendor's page, our own test runs, and where each tool fails12 tools / 36 pagesOct 30NEW
Glossary →AI terms, each defined in a sentence and tied to what it costs on live rates38 termsOct 21NEW
The Journal →Editorial: news cycle, deep analysis, opinion, research digests21 postsNov 18LIVE

Flagship rates

6 LABS · CURRENT FRONTIER TIER · RATES VERIFIED 2026-11-08 All models →

[

Anthropic

Claude Fable 5.2

$8/M

Input

$40/M

Output

$5.01/M

Blended

1M ctx · cache $0.20/M

SEE DETAILS →

](https://aicost.tools/llm-cost/anthropic/claude-fable-5-2/)[

OpenAI

GPT-6.2 Astra

$8/M

Input

$40/M

Output

$5.16/M

Blended

1.05M ctx · cache $0.40/M

SEE DETAILS →

](https://aicost.tools/llm-cost/openai/gpt-6-2-astra/)[

Google DeepMind

Gemini 4 Carbon

$2/M

Input

$10/M

Output

$1.21/M

Blended

1M ctx · cache $0.10/M

SEE DETAILS →

](https://aicost.tools/llm-cost/google/gemini-4-carbon/)[

SpaceXSI

Grok 4.7

$2/M

Input

$6/M

Output

$1.19/M

Blended

500K ctx · cache $0.50/M

SEE DETAILS →

](https://aicost.tools/llm-cost/xai/grok-4-7/)[

Moonshot (Kimi)

Kimi K3.1 Pro

$2/M

Input

$10/M

Output

$1.28/M

Blended

1M ctx · cache $0.20/M

SEE DETAILS →

](https://aicost.tools/llm-cost/moonshot/kimi-k3-1-pro/)[CHEAPEST

Zhipu (Z.ai / GLM)

GLM-5.3

$1.40/M

Input

$4.40/M

Output

$0.78/M

Blended

1M ctx · cache $0.26/M

SEE DETAILS →

](https://aicost.tools/llm-cost/zhipu/glm-5-3/)

Blended rate assumes a 92% input / 8% output, 82% cached workload — the same assumption behind every effective-rate figure on this site. Raw input and output rates are read off each vendor's own pricing page. Where a vendor publishes no cached-input rate, no cache discount is applied, so that model's blended figure is a ceiling rather than a like-for-like number.

Featured comparisons

SOURCE-VERIFIED · LIVE PRICING & BENCHMARKS All comparisons →

[

§ COMPARE / 012026-11

KimivsQwen

Moonshot and Alibaba both publish model weights and both sell plans on top of the API, but they package them…

Live pricing, benchmarks, and a verdict by use case.

](https://aicost.tools/compare/kimi-vs-qwen/)[

§ COMPARE / 022026-11

ChatGPT PlusvsClaude Pro

ChatGPT Plus and Claude Pro both cost $20 a month. Only Claude Pro can be paid annually, at $17 a month for…

Live pricing, benchmarks, and a verdict by use case.

](https://aicost.tools/compare/chatgpt-plus-vs-claude-pro/)[

§ COMPARE / 032026-11

ChatGPT ProvsClaude Max

The two power plans meet at two prices: ChatGPT Pro 100 against Claude Max at $100 a month, and Pro 200…

Live pricing, benchmarks, and a verdict by use case.

](https://aicost.tools/compare/chatgpt-pro-vs-claude-max/)[

§ COMPARE / 042026-11

Google AI ProvsChatGPT Plus

Google AI Pro is $19.99 a month, or $16.67 on the yearly plan; ChatGPT Plus is $20, monthly only. AI Pro…

Live pricing, benchmarks, and a verdict by use case.

](https://aicost.tools/compare/google-ai-pro-vs-chatgpt-plus/)[

§ COMPARE / 052026-11

Perplexity ProvsClaude Pro

On annual billing the two cost the same: Perplexity Pro $17 a month and Claude Pro $17. Perplexity is a search…

Live pricing, benchmarks, and a verdict by use case.

](https://aicost.tools/compare/perplexity-pro-vs-claude-pro/)[

§ COMPARE / 062026-11

Claude MaxvsUltra

Max is $100 a month for 5x Pro's usage every five hours and every week. Ultra is $200 for 10x Pro's weekly…

Live pricing, benchmarks, and a verdict by use case.

](https://aicost.tools/compare/claude-max-vs-ultra/)

§ 03 / JOURNAL

Latest from the desk.

All entries →

ANALYSISNOV 18 · 12 MIN

Compare LLM costs per task, not per token: what the measured bills show

cost-optimizationmodel-selectionapi-pricing

↗

TOOLS · HOW-TOSEP 27 · 17 MIN

input_tokens means three different things. Your cost function knows one of them.

openaianthropicgemini

↗

TOOLS · HOW-TO · FEATUREDSEP 19 · 17 MIN

A 429 is free. The ninth retry is not.

rate-limitsapi-errorsprompt-caching

↗

ANALYSIS · FEATUREDAUG 31 · 24 MIN

The list price is not the price your agent pays

prompt-cachingcost-optimizationagents

↗

ANALYSIS · FEATUREDAUG 27 · 22 MIN

Running many agents at once: the tax is linear, the rebate is quadratic

claude-codeagentssubagents

↗

TOOLS · HOW-TO · FEATUREDAUG 23 · 19 MIN

The 20x in Claude Code caching is real — it's a penalty, not a discount

claude-codeprompt-cachinganthropic

↗

The tape doesn't lie.

Every price move, model release, and policy change in the AI market — logged within hours, not weeks.

NOV 20

Mistral Large 4 repriced — output now $2.1/M tokens.

−49.8%

NOV 18

Gemini 4 Neon launched — $3 / $15 per M tokens.

RELEASE

NOV 17

GPT-6.2 Astra launched — $8 / $40 per M tokens.

RELEASE

NOV 17

GPT-6.2 Sol launched — $2 / $10 per M tokens.

RELEASE

NOV 16

Qwen4 Max launched — $1.5 / $4.5 per M tokens.

RELEASE

NOV 12

Gemini 4 Hydrogen launched — $1.5 / $7.5 per M tokens.

RELEASE

§ 04 / DESK NOTES

How we actually work.

Full methodology →

01

Vendor-canonical, not press releases

Pricing data is verified manually against each vendor's own canonical pricing page — 23 frontier providers, no scraping, no third-party aggregators. Every number on this site has a verification date stamped next to it.

02

One verified dataset behind every page

Model pricing pages, provider hubs, comparisons and calculators all read the same snapshot of vendor rates. A section goes live only when its data is checked, so there are no placeholder pages.

03

No silent edits. Ever.

When a number is wrong, the page is fixed and the correction goes into a public log: date, page, what was wrong and what is right now.

Independent operator. Founder verifies every claim before publish. Real LinkedIn linked from author profile.

Founder bio →