Skip to content
AuraMatrix

Token Price Index

What LLM tokens cost, channel by channel.

List prices for model APIs from vendors, clouds and aggregators. Each number links to the page it came from, with the time it was observed and a hash of what we saw.

Last sweep 2026-11-15 22:15 UTC123 of 123 prices confirmed against their sources24 models22 channelsUSD per 1M tokens

2026-11-15DeepSeek V4.1 Flash is 83% cheaper on OpenRouter0.30 → 0.0507

AuraMatrix token price indicestrialOfficial list prices by tier, 2026-11-03 = 100. How it is built · JSON

Flagship12 models100.0unchanged since the starttypical price $5.3313 per 1M tokens (3 parts input, 1 part output)Announced: 105.9 from 2026-12-01 (+5.9%)
Mid-range4 models100.0unchanged since the starttypical price $2.6321 per 1M tokens (3 parts input, 1 part output)Announced: 118.9 from 2027-01-01 (+18.9%)
Small and fast7 models100.0unchanged since the starttypical price $0.3068 per 1M tokens (3 parts input, 1 part output)No announced change ahead

GPU-hour reference pricestrialMedian on-demand price for one GPU for one hour across providers. How it is built

Moved in the last 7 days

Every change →

Announced, not yet in force

Use in your AIAsk your assistant instead of looking it up. One click for Cursor and VS Code.Connect

Use in your AI

Connect once, then just ask: "Where is Claude Opus 5.5 cheapest?", "What does an H100 cost per hour?", "Which models got cheaper this week?" Your assistant answers from this index, with the source for every number. Free, read-only, no account.

How it works →
{
  "mcpServers": {
    "auramatrix": {
      "url": "https://auramatrix.ai/mcp"
    }
  }
}
Add to CursorOne click opens Cursor and asks you to confirm. Or paste this into its MCP settings.

24 of 24 models

Official list price per model, and the lowest input price seen on any channel. USD per 1M tokens.
ModelVendorInputofficialOutputofficialLowest inputany channelChannelsOfficial status
Qwen3.8 Maxalibaba/qwen3.8-maxAlibaba (Qwen)2.006.001.65Model Studio, China (Beijing) · −18%4VERIFIED
Qwen3.8 Flashalibaba/qwen3.8-flashAlibaba (Qwen)0.150.470.113Model Studio, China (Beijing) · −25%4VERIFIED
Claude Fable 5.1anthropic/claude-fable-5.1Anthropic10.0050.0010.00official is lowest2VERIFIED
Claude Fable 5.2anthropic/claude-fable-5.2Anthropic8.0040.008.00official is lowest1VERIFIED
Claude Mythos 5.2anthropic/claude-mythos-5.2Anthropic8.0040.008.00official is lowest1VERIFIED
Claude Opus 5.5anthropic/claude-opus-5.5Anthropic4.0020.004.00official is lowest2VERIFIED
Claude Sonnet 5.5anthropic/claude-sonnet-5.5Anthropic2.0010.002.00official is lowest2VERIFIED
Claude Haiku 5.5anthropic/claude-haiku-5.5Anthropic0.100.500.10official is lowest2VERIFIED
DeepSeek V4 Prodeepseek/deepseek-v4-proDeepSeek1.323.960.90OpenRouter · −32%3VERIFIED
DeepSeek V4.1 Flashdeepseek/deepseek-v4.1-flashDeepSeek0.301.200.0507OpenRouter · −83%3VERIFIED
Gemini 4 Argongoogle/gemini-4-argonGoogle2.0010.002.00official is lowest4VERIFIED
Gemini 3.8 Flashgoogle/gemini-3.8-flashGoogle0.753.750.75official is lowest4VERIFIED
Gemini 3.5 Flash-Litegoogle/gemini-3.5-flash-liteGoogle0.302.500.30official is lowest4VERIFIED
Llama 4 Maverickmeta/llama-4-maverickMeta——0.24Amazon Bedrock, US East (N. Virginia)2no official price
MiniMax M3minimax/minimax-m3MiniMax0.301.200.30official is lowest3VERIFIED
MiniMax M3.1 Flashminimax/minimax-m3.1-flashMiniMax0.150.600.075OpenRouter · −50%3VERIFIED
Kimi K3moonshot/kimi-k3Moonshot AI (Kimi)3.0015.003.00official is lowest4VERIFIED
Kimi K3.1 Promoonshot/kimi-k3.1-proMoonshot AI (Kimi)2.0010.002.00official is lowest3VERIFIED
GPT-6 Astraopenai/gpt-6-astraOpenAI10.0050.0010.00official is lowest5VERIFIED
GPT-6.1 Solopenai/gpt-6.1-solOpenAI2.0010.002.00official is lowest5VERIFIED
GPT-6.1 Terraopenai/gpt-6.1-terraOpenAI1.005.001.00official is lowest5VERIFIED
GPT-6 Lunaopenai/gpt-6-lunaOpenAI0.100.500.10official is lowest5VERIFIED
GLM-5.3zhipu/glm-5.3Zhipu AI (Z.ai)1.404.400.084OpenRouter · −94%3VERIFIED
GLM-5.3-Flashzhipu/glm-5.3-flashZhipu AI (Z.ai)0.150.500.0399OpenRouter · −73%3VERIFIED

Cheapest in class

How to read this

Tiers follow each vendor's own positioning of a model within its lineup (flagship, mid-range, small/fast). They are an editorial grouping for comparing like with like, not a capability ranking. Ranked by official input price per 1M tokens, with output price alongside. Rank changes will appear once we have history.

Flagship

Each vendor's top model.

  1. 1MiniMax M30.30 / 1.20
  2. 2DeepSeek V4 Pro1.32 / 3.96
  3. 3GLM-5.31.40 / 4.40
  4. 4Qwen3.8 Max2.00 / 6.00
  5. 5Gemini 4 Argon2.00 / 10.00
  6. 6Kimi K3.1 Pro2.00 / 10.00
  7. 7Kimi K33.00 / 15.00
  8. 8Claude Opus 5.54.00 / 20.00
  9. 9Claude Fable 5.28.00 / 40.00
  10. 10Claude Mythos 5.28.00 / 40.00
  11. 11Claude Fable 5.110.00 / 50.00
  12. 12GPT-6 Astra10.00 / 50.00

Mid-range

The balanced model in the lineup.

  1. 1Gemini 3.8 Flash0.75 / 3.75
  2. 2GPT-6.1 Terra1.00 / 5.00
  3. 3Claude Sonnet 5.52.00 / 10.00
  4. 4GPT-6.1 Sol2.00 / 10.00
  5. –Llama 4 Maverick0.24 on AWS

Small and fast

The cheapest model each vendor pitches for high volume.

  1. 1Claude Haiku 5.50.10 / 0.50
  2. 2GPT-6 Luna0.10 / 0.50
  3. 3Qwen3.8 Flash0.15 / 0.47
  4. 4GLM-5.3-Flash0.15 / 0.50
  5. 5MiniMax M3.1 Flash0.15 / 0.60
  6. 6DeepSeek V4.1 Flash0.30 / 1.20
  7. 7Gemini 3.5 Flash-Lite0.30 / 2.50

Price ladder

How to read this

Official list price per 1M tokens, log scale. Input is the solid bar, output is the lighter one. Models without an official price are not shown.

  1. Claude Fable 5.110.0050.00
  2. GPT-6 Astra10.0050.00
  3. Claude Fable 5.28.0040.00
  4. Claude Mythos 5.28.0040.00
  5. Claude Opus 5.54.0020.00
  6. Kimi K33.0015.00
  7. Qwen3.8 Max2.006.00
  8. Claude Sonnet 5.52.0010.00
  9. Gemini 4 Argon2.0010.00
  10. Kimi K3.1 Pro2.0010.00
  11. GPT-6.1 Sol2.0010.00
  12. GLM-5.31.404.40
  13. DeepSeek V4 Pro1.323.96
  14. GPT-6.1 Terra1.005.00
  15. Gemini 3.8 Flash0.753.75
  16. DeepSeek V4.1 Flash0.301.20
  17. Gemini 3.5 Flash-Lite0.302.50
  18. MiniMax M30.301.20
  19. Qwen3.8 Flash0.150.47
  20. MiniMax M3.1 Flash0.150.60
  21. GLM-5.3-Flash0.150.50
  22. Claude Haiku 5.50.100.50
  23. GPT-6 Luna0.100.50

Same model, different region

How to read this

Where an operator publishes different prices by region or deployment. Input price per 1M tokens; the cheapest region is the baseline.

Qwen3.8 MaxAlibaba Cloud

  • CNChinaChina (Beijing)1.65base
  • SGSingaporeInternational (Singapore)2.00+21%

Qwen3.8 FlashAlibaba Cloud

  • CNChinaChina (Beijing)0.113base
  • SGSingaporeInternational (Singapore)0.15+33%

Gemini 4 ArgonGoogle Cloud

  • globalGlobal endpointglobal endpoint4.00base
  • regionalRegional endpointsregional endpoint (any pinned region)4.40+10%

Gemini 3.8 FlashGoogle Cloud

  • globalGlobal endpointglobal endpoint1.50base
  • regionalRegional endpointsregional endpoint (any pinned region)1.65+10%

Gemini 3.5 Flash-LiteGoogle Cloud

  • globalGlobal endpointglobal endpoint0.30base
  • regionalRegional endpointsregional endpoint (any pinned region)0.33+10%

Kimi K3AWS

  • globalGlobal endpointglobal cross-region3.00base
  • USUnited StatesUS East (N. Virginia)3.30+10%

GPT-6 AstraMicrosoft Azure

  • globalGlobal endpointGlobal deployment10.00base
  • USUnited StatesData Zone (US)11.00+10%
  • EUEuropean UnionData Zone (EU)12.00+20%

GPT-6.1 SolMicrosoft Azure

  • globalGlobal endpointGlobal deployment2.00base
  • USUnited StatesData Zone (US)2.20+10%
  • EUEuropean UnionData Zone (EU)2.40+20%

GPT-6.1 TerraMicrosoft Azure

  • globalGlobal endpointGlobal deployment1.00base
  • USUnited StatesData Zone (US)1.10+10%
  • EUEuropean UnionData Zone (EU)1.20+20%

GPT-6 LunaMicrosoft Azure

  • globalGlobal endpointGlobal deployment0.10base
  • USUnited StatesData Zone (US)0.11+10%
  • EUEuropean UnionData Zone (EU)0.12+20%

One job, every model

How to read this

Cost of sending 1,000,000 input tokens and receiving 100,000 output tokens at each model's official list price. Cached, batch and tier rates are not applied. Each vendor counts tokens with its own tokenizer, so the same text is not the same number of tokens everywhere; this compares prices per token, not the cost of a given document.

  1. Claude Haiku 5.5$0.15
  2. GPT-6 Luna$0.15
  3. Qwen3.8 Flash$0.20
  4. GLM-5.3-Flash$0.20
  5. MiniMax M3.1 Flash$0.21
  6. DeepSeek V4.1 Flash$0.42
  7. MiniMax M3$0.42
  8. Gemini 3.5 Flash-Lite$0.55
  9. Gemini 3.8 Flash$1.13
  10. GPT-6.1 Terra$1.50
  11. DeepSeek V4 Pro$1.72
  12. GLM-5.3$1.84
  13. Qwen3.8 Max$2.60
  14. Claude Sonnet 5.5$3.00
  15. Gemini 4 Argon$3.00
  16. Kimi K3.1 Pro$3.00
  17. GPT-6.1 Sol$3.00
  18. Kimi K3$4.50
  19. Claude Opus 5.5$6.00
  20. Claude Fable 5.2$12
  21. Claude Mythos 5.2$12
  22. Claude Fable 5.1$15
  23. GPT-6 Astra$15

Who moves the most tokens on OpenRouter

Share of all tokens routed through OpenRouter on 2026-11-03 (165.3T tokens that day), as published on its rankings page. This is one marketplace's traffic, weighted toward open-weight and low-price models; traffic that goes straight to vendor APIs is not visible to anyone.

  1. DeepSeek V4.1 FlashDeepSeek42.6%70.4T
  2. GLM-5.3-FlashZhipu AI9.9%16.4T
  3. hy4-previewtencent5.7%9.5T
  4. GPT-6 LunaOpenAI5.6%9.3T
  5. step-5-previewstepfun4.0%6.7T
  6. Claude Opus 5.5Anthropic4.0%6.7T
  7. MiniMax M3.1 FlashMiniMax3.9%6.5T
  8. gpt-5.6-lunaOpenAI3.2%5.3T
  9. GPT-6.1 SolOpenAI2.5%4.0T
  10. deepseek-v4-flashDeepSeek2.3%3.8T
  11. GLM-5.3Zhipu AI2.2%3.7T
  12. jev-1.13typesafe1.9%3.2T

Source: OpenRouter rankings page (embedded daily model stats), fetched 2026-11-04 05:46 UTC, snapshot SHA-256 ce75bb289e41396a…. Models without a link are not in the price index yet.

What changed

Price changes seen between successive records from the same source, and changes vendors have announced with a date. Recording began 2026-11-03; every record stays, so this list only grows. JSON feed.

  1. 2027-01-01ANNOUNCEDGemini 3.8 FlashGemini Developer APIin 0.75 → 1.50 · out 3.75 → 7.50 from 2027-01-01; current price valid until 2026-12-31
  2. 2027-01-01ANNOUNCEDGemini 3.8 FlashGemini Developer API · Batchin 0.375 → 0.75 · out 1.875 → 3.75 from 2027-01-01; current price valid until 2026-12-31
  3. 2026-12-01ANNOUNCEDGemini 4 ArgonGemini Developer APIin 2.00 → 4.00 · out 10.00 → 20.00 from 2026-12-01; current price valid until 2026-11-30
  4. 2026-12-01ANNOUNCEDGemini 4 ArgonGemini Developer API · Batchin 1.00 → 2.00 · out 5.00 → 10.00 from 2026-12-01; current price valid until 2026-11-30
  5. 2026-11-15OBSERVEDDeepSeek V4.1 FlashOpenRouterin 0.30 → 0.0507 · out 1.20 → 0.2028 observed 2026-11-15 22:15 UTC
  6. 2026-11-13OBSERVEDDeepSeek V4 ProOpenRouterin 0.889 → 0.90 · out 1.7779 → 1.50 observed 2026-11-13 22:15 UTC
  7. 2026-11-13OBSERVEDKimi K3.1 ProOpenRouterin 0.2375 → 2.00 · out 8.4114 → 10.00 observed 2026-11-13 22:15 UTC
  8. 2026-11-13OBSERVEDGLM-5.3OpenRouterin 0.0699 → 0.084 · out 1.40 → 3.60 observed 2026-11-13 22:15 UTC
  9. 2026-11-12OBSERVEDDeepSeek V4 ProOpenRouterin 0.8754 → 0.889 · out 1.7508 → 1.7779 observed 2026-11-12 22:15 UTC
  10. 2026-11-12OBSERVEDDeepSeek V4.1 FlashOpenRouterin 0.051 → 0.30 · out 0.567 → 1.20 observed 2026-11-12 22:15 UTC
  11. 2026-11-12OBSERVEDKimi K3.1 ProOpenRouterin 2.00 → 0.2375 · out 10.00 → 8.4114 observed 2026-11-12 22:15 UTC
  12. 2026-11-11OBSERVEDDeepSeek V4 ProOpenRouterin 0.889 → 0.8754 · out 1.7779 → 1.7508 observed 2026-11-11 22:15 UTC

Email me when a price changes

One email after our daily check if a price changed; nothing on quiet days. Confirm via the link we send you; unsubscribe in one click.

19 more in the feed.

About these prices: units, what the status marks mean, corrections

Units

US dollars per 1 million tokens, as published by each channel. China-site price lists are shown in CNY, as published; nothing is converted. Last sweep 2026-11-15 22:15 UTC.

Trust status

23 of 24 official list prices have been re-read by an independent second reader so far: a separate AI model reads the snapshot without seeing the recorded number, and humans spot-check.

  • Verified: read from the operator's own price page, and an independent second reader got the same numbers from the snapshot.
  • Listed: a marketplace listing, re-read the same way; what a marketplace bills can depend on where it routes, so it is never the reference.
  • Unverified: recorded, not yet re-read.

Not advice

Unofficial and not affiliated with any vendor. Vendor names are trademarks of their owners. The channel's own page is authoritative. Missing prices are unknown, not free.

Data

The whole dataset as JSON, the same records as these pages. Corrections: hello@auramatrix.ai