# TokenMargin.dev — Live 2026 LLM API & Cloud GPU Pricing Table (Markdown for AI Agents)

> **Zero-Auth Agent-Ready Pricing Feed (`text/markdown`)**
> - **Canonical URL:** `https://tokenmargin.dev/pricing.md`
> - **Streamable HTTP MCP Server:** `https://tokenmargin.dev/mcp` (Discovery: `https://tokenmargin.dev/.well-known/mcp.json`)
> - **Structured JSON API:** `https://tokenmargin.dev/api/v1/pricing.json`
> - **AI Answer Engine Index:** `https://tokenmargin.dev/llms.txt`
> - **Last Verified Sync:** `2026-11-19T01:13:20.817Z` (480 OpenRouter models · 324 inference providers · 134 curated production LLMs)

## 1. All 134 Curated Production LLM API Prices (USD per 1M Tokens)

| Model | ID | Provider | Tier | Input ($/1M) | Output ($/1M) | Cache Read ($/1M) | Context | Est. P50 TTFT | Speed (tok/s) | 10K DAU Bill (Uncached) | 10K DAU (75% Cache) |
| :--- | :--- | :--- | :--- | ---: | ---: | ---: | ---: | ---: | ---: | ---: | ---: |
| **Claude Sonnet 4** | `anthropic/claude-sonnet-4` | Anthropic | frontier | $3.00 | $15.00 | $0.30 | 200K | 390ms | 84 | $81,000/mo | $55,485/mo |
| **Claude 3.7 Sonnet** | `anthropic/claude-3.7-sonnet` | Anthropic | frontier | $3.00 | $15.00 | $0.30 | 200K | 420ms | 78 | $81,000/mo | $55,485/mo |
| **Claude 3.7 Sonnet (Extended Thinking)** | `anthropic/claude-3.7-sonnet:thinking` | Anthropic | reasoning | $3.00 | $15.00 | $0.30 | 200K | 650ms | 68 | $81,000/mo | $55,485/mo |
| **Claude 3.5 Sonnet (v2)** | `anthropic/claude-3.5-sonnet` | Anthropic | frontier | $3.00 | $15.00 | $0.30 | 200K | 390ms | 82 | $81,000/mo | $55,485/mo |
| **Claude 3.5 Haiku** | `anthropic/claude-3.5-haiku` | Anthropic | fast | $0.80 | $4.00 | $0.08 | 200K | 240ms | 135 | $21,600/mo | $14,796/mo |
| **Claude 3 Opus (Legacy High-Cost)** | `anthropic/claude-3-opus` | Anthropic | frontier | $15.00 | $75.00 | $1.50 | 200K | 890ms | 32 | $405,000/mo | $277,425/mo |
| **GPT-4o** | `openai/gpt-4o` | OpenAI | frontier | $2.50 | $10.00 | $1.25 | 128K | 320ms | 98 | $60,300/mo | $48,488/mo |
| **GPT-4o Mini** | `openai/gpt-4o-mini` | OpenAI | fast | $0.15 | $0.60 | $0.075 | 128K | 210ms | 148 | $3,618/mo | $2,909/mo |
| **OpenAI o3 (Frontier Reasoning)** | `openai/o3` | OpenAI | reasoning | $2.00 | $8.00 | $0.50 | 200K | 480ms | 96 | $48,240/mo | $34,065/mo |
| **OpenAI o3-mini** | `openai/o3-mini` | OpenAI | reasoning | $1.10 | $4.40 | $0.55 | 200K | 520ms | 118 | $26,532/mo | $21,335/mo |
| **OpenAI o1 (Frontier Reasoning)** | `openai/o1` | OpenAI | reasoning | $15.00 | $60.00 | $7.50 | 200K | 950ms | 62 | $361,800/mo | $290,925/mo |
| **GPT-4.5 Preview (Cost Trap Alert)** | `openai/gpt-4.5-preview` | OpenAI | frontier | $75.00 | $150.00 | $37.50 | 128K | 780ms | 45 | $1,377,000/mo | $1,022,625/mo |
| **GPT-4 Turbo (Legacy)** | `openai/gpt-4-turbo` | OpenAI | frontier | $10.00 | $30.00 | $5.00 | 128K | 560ms | 48 | $212,400/mo | $165,150/mo |
| **DeepSeek V3 (0324)** | `deepseek/deepseek-chat-v3-0324` | DeepInfra | open-weight | $0.24 | $0.79 | $0.07 | 164K | 360ms | 92 | $5,299/mo | $3,693/mo |
| **DeepSeek R1 (671B Reasoning)** | `deepseek/deepseek-r1` | DeepInfra | reasoning | $0.50 | $2.15 | $0.14 | 164K | 490ms | 76 | $12,492/mo | $9,090/mo |
| **DeepSeek R1 Distill Llama 70B** | `deepseek/deepseek-r1-distill-llama-70b` | Groq | reasoning | $0.23 | $0.69 | $0.11 | 131K | 140ms | 390 | $4,885/mo | $3,751/mo |
| **DeepSeek R1 Distill Qwen 32B** | `deepseek/deepseek-r1-distill-qwen-32b` | DeepInfra | reasoning | $0.12 | $0.18 | $0.06 | 131K | 210ms | 140 | $2,030/mo | $1,463/mo |
| **Gemini 2.5 Pro** | `google/gemini-2.5-pro-preview-03-25` | Google | frontier | $1.25 | $10.00 | $0.31 | 1.0M | 380ms | 112 | $44,550/mo | $35,667/mo |
| **Gemini 2.5 Flash** | `google/gemini-2.5-flash` | Google | fast | $0.30 | $2.50 | $0.03 | 1.0M | 175ms | 215 | $10,980/mo | $8,429/mo |
| **Gemini 2.0 Flash** | `google/gemini-2.0-flash-001` | Google | fast | $0.10 | $0.40 | $0.025 | 1.0M | 185ms | 195 | $2,412/mo | $1,703/mo |
| **Gemini 2.0 Flash-Lite** | `google/gemini-2.0-flash-lite-001` | Google | fast | $0.075 | $0.30 | $0.019 | 1.0M | 150ms | 245 | $1,809/mo | $1,280/mo |
| **Gemini 1.5 Pro (2M Context)** | `google/gemini-1.5-pro` | Google | frontier | $1.25 | $5.00 | $0.313 | 2.1M | 360ms | 88 | $30,150/mo | $21,291/mo |
| **Gemini 1.5 Flash** | `google/gemini-1.5-flash` | Google | fast | $0.075 | $0.30 | $0.019 | 1.0M | 195ms | 185 | $1,809/mo | $1,277/mo |
| **Llama 3.3 70B Instruct** | `meta-llama/llama-3.3-70b-instruct` | DeepInfra | open-weight | $0.12 | $0.30 | $0.06 | 131K | 240ms | 115 | $2,376/mo | $1,809/mo |
| **Llama 3.1 405B Instruct** | `meta-llama/llama-3.1-405b-instruct` | DeepInfra | open-weight | $0.80 | $0.80 | $0.40 | 131K | 430ms | 64 | $12,384/mo | $8,604/mo |
| **Llama 3.1 8B Instruct** | `meta-llama/llama-3.1-8b-instruct` | Groq | fast | $0.03 | $0.05 | $0.015 | 131K | 90ms | 750 | $522.00/mo | $380.25/mo |
| **Llama 3.2 90B Vision Instruct** | `meta-llama/llama-3.2-90b-vision-instruct` | DeepInfra | open-weight | $0.35 | $0.40 | $0.18 | 131K | 310ms | 96 | $5,562/mo | $3,956/mo |
| **Llama 3.2 11B Vision Instruct** | `meta-llama/llama-3.2-11b-vision-instruct` | DeepInfra | fast | $0.049 | $0.049 | $0.025 | 131K | 165ms | 210 | $758.52/mo | $531.72/mo |
| **Llama 3.2 3B Instruct (Edge/Router)** | `meta-llama/llama-3.2-3b-instruct` | DeepInfra | fast | $0.015 | $0.025 | $0.0080 | 131K | 85ms | 480 | $261.00/mo | $194.85/mo |
| **Qwen QwQ-32B (Reasoning)** | `qwen/qwq-32b` | DeepInfra | reasoning | $0.12 | $0.18 | $0.06 | 131K | 210ms | 155 | $2,030/mo | $1,463/mo |
| **Qwen 2.5 72B Instruct** | `qwen/qwen-2.5-72b-instruct` | DeepInfra | open-weight | $0.12 | $0.39 | $0.06 | 33K | 260ms | 105 | $2,635/mo | $2,068/mo |
| **Qwen 2.5 Coder 32B Instruct** | `qwen/qwen-2.5-coder-32b-instruct` | DeepInfra | open-weight | $0.07 | $0.16 | $0.035 | 33K | 215ms | 142 | $1,343/mo | $1,012/mo |
| **Qwen 2.5 7B Instruct** | `qwen/qwen-2.5-7b-instruct` | DeepInfra | fast | $0.04 | $0.10 | $0.02 | 131K | 130ms | 260 | $792.00/mo | $603.00/mo |
| **Qwen Turbo (1M Context)** | `qwen/qwen-turbo` | OpenRouter | fast | $0.05 | $0.20 | $0.025 | 1M | 170ms | 220 | $1,206/mo | $969.75/mo |
| **Mistral Large 2 (2411)** | `mistralai/mistral-large-2411` | OpenRouter | frontier | $2.00 | $6.00 | $1.00 | 131K | 340ms | 86 | $42,480/mo | $33,030/mo |
| **Mistral Small 3.1 24B Instruct** | `mistralai/mistral-small-3.1-24b-instruct` | DeepInfra | open-weight | $0.05 | $0.10 | $0.025 | 128K | 180ms | 175 | $918.00/mo | $681.75/mo |
| **Mistral Codestral 2501** | `mistralai/codestral-2501` | OpenRouter | fast | $0.30 | $0.90 | $0.15 | 262K | 210ms | 165 | $6,372/mo | $4,955/mo |
| **Ministral 8B** | `mistralai/ministral-8b` | OpenRouter | fast | $0.10 | $0.10 | $0.05 | 131K | 155ms | 210 | $1,548/mo | $1,076/mo |
| **xAI Grok 2 (1212)** | `x-ai/grok-2-1212` | OpenRouter | frontier | $2.00 | $10.00 | $0.50 | 131K | 330ms | 94 | $54,000/mo | $39,825/mo |
| **xAI Grok 2 Vision** | `x-ai/grok-2-vision-1212` | OpenRouter | frontier | $2.00 | $10.00 | $0.50 | 33K | 360ms | 88 | $54,000/mo | $39,825/mo |
| **Cohere Command A (Enterprise RAG)** | `cohere/command-a` | OpenRouter | frontier | $2.50 | $10.00 | $1.25 | 256K | 290ms | 110 | $60,300/mo | $48,488/mo |
| **Cohere Command R+ (08-2024)** | `cohere/command-r-plus-08-2024` | OpenRouter | frontier | $2.50 | $10.00 | $1.25 | 128K | 350ms | 84 | $60,300/mo | $48,488/mo |
| **Cohere Command R7B (Fast RAG)** | `cohere/command-r7b-12-2024` | OpenRouter | fast | $0.037 | $0.15 | $0.019 | 128K | 145ms | 230 | $904.50/mo | $729.68/mo |
| **Amazon Nova Pro v1** | `amazon/nova-pro-v1` | OpenRouter | frontier | $0.80 | $3.20 | $0.20 | 300K | 280ms | 118 | $19,296/mo | $13,626/mo |
| **Amazon Nova Lite v1** | `amazon/nova-lite-v1` | OpenRouter | fast | $0.06 | $0.24 | $0.015 | 300K | 160ms | 225 | $1,447/mo | $1,022/mo |
| **Amazon Nova Micro v1** | `amazon/nova-micro-v1` | OpenRouter | fast | $0.035 | $0.14 | $0.0088 | 128K | 120ms | 290 | $844.20/mo | $596.61/mo |
| **Microsoft Phi-4 (14B Dense)** | `microsoft/phi-4` | DeepInfra | open-weight | $0.06 | $0.14 | $0.03 | 16K | 145ms | 195 | $1,159/mo | $875.70/mo |
| **NVIDIA Nemotron 70B Instruct** | `nvidia/llama-3.1-nemotron-70b-instruct` | DeepInfra | open-weight | $0.12 | $0.30 | $0.06 | 131K | 250ms | 108 | $2,376/mo | $1,809/mo |
| **Google Gemma 3 27B Instruct** | `google/gemma-3-27b-it` | DeepInfra | open-weight | $0.10 | $0.20 | $0.05 | 131K | 190ms | 145 | $1,836/mo | $1,364/mo |
| **Perplexity Sonar Reasoning Pro** | `perplexity/sonar-reasoning-pro` | OpenRouter | reasoning | $2.00 | $8.00 | $1.00 | 128K | 410ms | 95 | $48,240/mo | $38,790/mo |
| **GPT-4 32k (Deprecated Legacy)** | `openai/gpt-4-32k` | OpenAI | frontier | $60.00 | $120.00 | $30.00 | 33K | 920ms | 22 | $1,101,600/mo | $818,100/mo |
| **OpenAI o1-mini** | `openai/o1-mini` | OpenAI | reasoning | $1.10 | $4.40 | $0.55 | 128K | 490ms | 112 | $26,532/mo | $21,335/mo |
| **GPT-4.1 (1M Context Frontier)** | `openai/gpt-4.1` | OpenAI | frontier | $2.00 | $8.00 | $0.50 | 1.0M | 310ms | 106 | $48,240/mo | $34,065/mo |
| **GPT-4.1 Mini (75% Cache Discount)** | `openai/gpt-4.1-mini` | OpenAI | fast | $0.40 | $1.60 | $0.10 | 1.0M | 215ms | 145 | $9,648/mo | $6,813/mo |
| **GPT-4.1 Nano (Ultra-Fast Router)** | `openai/gpt-4.1-nano` | OpenAI | fast | $0.10 | $0.40 | $0.03 | 1.0M | 140ms | 210 | $2,412/mo | $1,751/mo |
| **Claude 3 Haiku** | `anthropic/claude-3-haiku` | Anthropic | fast | $0.25 | $1.25 | $0.03 | 200K | 190ms | 168 | $6,750/mo | $4,671/mo |
| **Llama 4 Maverick (17B-128E MoE)** | `meta-llama/llama-4-maverick` | DeepInfra | open-weight | $0.19 | $0.49 | $0.095 | 524K | 210ms | 148 | $3,805/mo | $2,907/mo |
| **Llama 4 Scout (10M Context MoE)** | `meta-llama/llama-4-scout` | Groq | open-weight | $0.08 | $0.30 | $0.04 | 1.3M | 110ms | 590 | $1,872/mo | $1,494/mo |
| **Qwen 2.5 Max (Frontier MoE)** | `qwen/qwen-max` | OpenRouter | frontier | $1.60 | $6.40 | $0.80 | 131K | 340ms | 88 | $38,592/mo | $31,032/mo |
| **Mistral Pixtral Large (124B Vision)** | `mistralai/pixtral-large-2411` | OpenRouter | frontier | $2.00 | $6.00 | $1.00 | 131K | 360ms | 78 | $42,480/mo | $33,030/mo |
| **DeepSeek R1 Distill Llama 8B** | `deepseek/deepseek-r1-distill-llama-8b` | DeepInfra | fast | $0.04 | $0.04 | $0.02 | 131K | 115ms | 340 | $619.20/mo | $430.20/mo |
| **Liquid AI d1.2-ultra** | `liquid/d1.2-ultra` | Liquid AI | fast | $0.042 | $0.0000 | $0.021 | 131K | 85ms | 420 | $529.20/mo | $330.75/mo |
| **Liquid AI d1.2** | `liquid/d1.2` | Liquid AI | fast | $0.02 | $0.0000 | $0.01 | 131K | 65ms | 510 | $252.00/mo | $157.50/mo |
| **Sage S2.1 Flash (320B/18B MoE)** | `shanghai-ai-lab/sage-s2.1-flash` | Shanghai AI Lab | fast | $0.06 | $0.24 | $0.015 | 1M | 115ms | 260 | $1,447/mo | $1,022/mo |
| **Kimi K3.1 Pro Abliterated (ZDR MXFP4)** | `moonshotai/kimi-k3.1-pro-abliterated` | Refuseless / Moonshot | frontier | $2.50 | $5.00 | $0.25 | 1M | 180ms | 145 | $45,900/mo | $24,638/mo |
| **OVHcloud Olala mini 7A1B (8GB VRAM / CPU)** | `ovhcloud/olala-mini-7a1b` | OVHcloud | open-weight | $0.03 | $0.09 | $0.0080 | 33K | 95ms | 160 | $637.20/mo | $429.30/mo |
| **Baidu ERNIE 5.5 Flash Preview (280B/14B · $0 Free)** | `baidu/ernie-5.5-flash-preview:free` | Baidu / OpenRouter | open-weight | $0.0000 | $0.0000 | $0.0000 | 1.0M | 160ms | 175 | $0.0000/mo | $0.0000/mo |
| **GLM 5.3 FlashX (200+ tok/s W8A8 High-Speed)** | `z-ai/glm-5.3-flashx` | Z.ai | fast | $0.37 | $1.25 | $0.075 | 1.0M | 110ms | 225 | $8,262/mo | $5,474/mo |
| **NVIDIA Nemotron 3 Ultra (550B / 55B MoE)** | `nvidia/nemotron-3-ultra-550b-a55b` | NVIDIA | open-weight | $0.50 | $2.20 | $0.10 | 262K | 190ms | 130 | $12,636/mo | $8,856/mo |
| **Microsoft MAI Code 2 (Ignite Promo -> 2x May 2027 Cliff)** | `microsoft/mai-code-2` | Microsoft Foundry | frontier | $0.50 | $2.50 | $0.025 | 1.1M | 185ms | 165 | $13,500/mo | $9,011/mo |
| **Microsoft MAI Code 2 Flash (1.05M Ctx · $0.05 Promo)** | `microsoft/mai-code-2-flash` | Microsoft Foundry | fast | $0.05 | $0.25 | $0.0050 | 1.1M | 95ms | 260 | $1,350/mo | $924.75/mo |
| **Microsoft MAI Decision 1.1 ($0.042 / $0 Output Router)** | `microsoft/decision-1.1` | Microsoft Foundry | fast | $0.042 | $0.0000 | $0.0080 | 131K | 42ms | 850 | $529.20/mo | $207.90/mo |
| **Tencent Hy4-Flash (224B/12B MoE · Open-Weight FP8)** | `tencent/hy4-flash` | Tencent / OpenRouter | open-weight | $0.035 | $0.141 | $0.0015 | 1.0M | 120ms | 155 | $845.64/mo | $529.07/mo |
| **Google Gemini 4 Neon (1.05M Ctx · $0.15 95% Cache)** | `google/gemini-4-neon` | Google Vertex AI / AI Studio | frontier | $3.00 | $15.00 | $0.15 | 1.0M | 145ms | 195 | $81,000/mo | $54,068/mo |
| **Mistral Large 4 "Le Chonk" (1.05T/49B MoE · $0 EU Vibe Promo)** | `mistralai/mistral-large-4-0` | Mistral AI (La Plateforme / Vibe) | open-weight | $0.70 | $2.10 | $0.07 | 1.0M | 175ms | 135 | $14,868/mo | $8,915/mo |
| **GPT-6.2 Astra Pro** | `openai/gpt-6.2-astra-pro` | OpenAI | open-weight | $8.00 | $40.00 | $0.40 | 1.1M | 440ms | 74 | $216,000/mo | $144,180/mo |
| **GPT-6.2 Astra** | `openai/gpt-6.2-astra` | OpenAI | open-weight | $8.00 | $40.00 | $0.40 | 1.1M | 440ms | 74 | $216,000/mo | $144,180/mo |
| **GPT-6.2 Sol Pro** | `openai/gpt-6.2-sol-pro` | OpenAI | open-weight | $2.00 | $10.00 | $0.10 | 1.1M | 440ms | 74 | $54,000/mo | $36,045/mo |
| **GPT-6.2 Sol** | `openai/gpt-6.2-sol` | OpenAI | open-weight | $2.00 | $10.00 | $0.10 | 1.1M | 310ms | 96 | $54,000/mo | $36,045/mo |
| **Qwen4 Max** | `qwen/qwen4-max` | Alibaba Qwen | open-weight | $1.43 | $4.28 | $0.071 | 1M | 310ms | 96 | $30,267/mo | $17,475/mo |
| **Gemini 4 Hydrogen** | `google/gemini-4-hydrogen` | Google | fast | $1.50 | $7.50 | $0.075 | 1.0M | 210ms | 155 | $40,500/mo | $27,034/mo |
| **ERNIE-Light-DT-1.0** | `baidu/ernie-light-dt-1.0-intl` | Baidu ERNIE | frontier | $1.13 | $3.94 | $0.113 | 1.0M | 310ms | 96 | $25,585/mo | $15,975/mo |
| **Gemini 4 Carbon** | `google/gemini-4-carbon` | Google | fast | $2.00 | $10.00 | $0.10 | 1.0M | 210ms | 155 | $54,000/mo | $36,045/mo |
| **Nex-N2.6** | `nex-agi/nex-n2.6` | Nex AGI | frontier | $0.12 | $0.40 | $0.024 | 262K | 310ms | 96 | $2,664/mo | $1,757/mo |
| **Inkling 2 Small** | `thinkingmachines/inkling-2-small` | Thinking Machines | fast | $0.30 | $1.00 | $0.03 | 1.0M | 210ms | 155 | $6,660/mo | $4,109/mo |
| **Claude Fable 5.2** | `anthropic/claude-fable-5.2` | Anthropic | reasoning | $8.00 | $40.00 | $0.20 | 1M | 440ms | 74 | $216,000/mo | $142,290/mo |
| **Muse Ember** | `meta/muse-ember` | Meta AI | fast | $0.10 | $0.50 | $0.0040 | 1.0M | 210ms | 155 | $2,700/mo | $1,793/mo |
| **LFM3-1.7B** | `liquid/lfm-3-1.7b` | Liquid AI | open-weight | $0.08 | $0.20 | $0.02 | 262K | 310ms | 96 | $1,584/mo | $1,017/mo |
| **Gemini Omni 1.2 Flash** | `google/gemini-omni-1.2-flash` | Google | fast | $1.50 | $9.00 | $0.375 | 131K | 210ms | 155 | $44,820/mo | $34,189/mo |
| **Hy4** | `tencent/hy4` | Tencent Hunyuan | frontier | $0.417 | $1.25 | $0.021 | 1.0M | 310ms | 96 | $8,854/mo | $5,112/mo |
| **GPT-6.1 Terra Pro** | `openai/gpt-6.1-terra-pro` | OpenAI | open-weight | $1.00 | $5.00 | $0.05 | 1.1M | 440ms | 74 | $27,000/mo | $18,023/mo |
| **GPT-6.1 Terra** | `openai/gpt-6.1-terra` | OpenAI | open-weight | $1.00 | $5.00 | $0.05 | 1.1M | 310ms | 96 | $27,000/mo | $18,023/mo |
| **Kimi K3.1 Pro** | `moonshotai/kimi-k3.1-pro` | Moonshot AI | open-weight | $0.253 | $8.80 | $0.18 | 1.0M | 440ms | 74 | $28,532/mo | $27,842/mo |
| **MiniMax M3.1 Flash** | `minimax/minimax-m3.1-flash` | MiniMax | open-weight | $0.15 | $0.60 | $0.015 | 1.0M | 210ms | 155 | $3,618/mo | $2,342/mo |
| **DeepSeek V4.1 Flash 1017** | `deepseek/deepseek-v4.1-flash-1017` | DeepSeek | open-weight | $0.30 | $1.20 | $0.0060 | 1.0M | 210ms | 155 | $7,236/mo | $4,458/mo |
| **Muse Spark Open** | `meta/muse-spark-open` | Meta AI | open-weight | $1.25 | $4.25 | $0.15 | 1.0M | 310ms | 96 | $27,990/mo | $17,595/mo |
| **Nemotron 3.5 Super VL** | `nvidia/nemotron-3.5-super-vl` | NVIDIA | open-weight | $0.15 | $0.50 | $0.02 | 1.0M | 310ms | 96 | $3,330/mo | $2,102/mo |
| **Gemini 4 Argon** | `google/gemini-4-argon` | Google | fast | $2.00 | $10.00 | $0.10 | 1.0M | 210ms | 155 | $54,000/mo | $36,045/mo |
| **Muse Wonder** | `meta/muse-wonder` | Meta AI | reasoning | $4.00 | $20.00 | $0.20 | 1.0M | 440ms | 74 | $108,000/mo | $72,090/mo |
| **Claude Haiku 5.5** | `anthropic/claude-haiku-5.5` | Anthropic | fast | $0.10 | $0.50 | $0.01 | 1M | 210ms | 155 | $2,700/mo | $1,850/mo |
| **Ling 3.1 Flash** | `inclusionai/ling-3.1-flash` | inclusionAI | fast | $0.045 | $0.135 | $0.0090 | 1M | 210ms | 155 | $955.80/mo | $615.60/mo |
| **GPT-6.1 Sol Pro** | `openai/gpt-6.1-sol-pro` | OpenAI | open-weight | $2.00 | $10.00 | $0.10 | 1.1M | 440ms | 74 | $54,000/mo | $36,045/mo |
| **GPT-6.1 Sol** | `openai/gpt-6.1-sol` | OpenAI | open-weight | $2.00 | $10.00 | $0.10 | 1.1M | 310ms | 96 | $54,000/mo | $36,045/mo |
| **Claude Sonnet 5.5** | `anthropic/claude-sonnet-5.5` | Anthropic | frontier | $2.00 | $10.00 | $0.10 | 1M | 310ms | 96 | $54,000/mo | $36,045/mo |
| **GLM 5.3 Prime** | `z-ai/glm-5.3-prime` | Z.ai (GLM) | open-weight | $1.68 | $5.28 | $0.336 | 1M | 310ms | 96 | $36,374/mo | $23,674/mo |
| **Qwen3.8 Max Prime** | `qwen/qwen3.8-max-prime` | Alibaba Qwen | open-weight | $4.00 | $12.00 | $0.50 | 1M | 310ms | 96 | $84,960/mo | $51,885/mo |
| **Command A+** | `cohere/command-a-plus` | Cohere | frontier | $0.30 | $1.50 | $0.15 | 192K | 310ms | 96 | $8,100/mo | $6,683/mo |
| **GPT-6 Luna Pro** | `openai/gpt-6-luna-pro` | OpenAI | open-weight | $0.10 | $0.50 | $0.01 | 1.1M | 210ms | 155 | $2,700/mo | $1,850/mo |
| **GPT-6 Luna** | `openai/gpt-6-luna` | OpenAI | open-weight | $0.10 | $0.50 | $0.01 | 1.1M | 210ms | 155 | $2,700/mo | $1,850/mo |
| **Claude Opus 5.5** | `anthropic/claude-opus-5.5` | Anthropic | frontier | $4.00 | $20.00 | $0.20 | 1M | 310ms | 96 | $108,000/mo | $72,090/mo |
| **MiMo-V2.6-Flash** | `xiaomi/mimo-v2.6-flash` | Xiaomi MiMo | fast | $0.14 | $0.28 | $0.0028 | 1.1M | 210ms | 155 | $2,570/mo | $1,274/mo |
| **MiMo-V2.6-Pro** | `xiaomi/mimo-v2.6-pro` | Xiaomi MiMo | reasoning | $0.435 | $0.87 | $0.0036 | 1.1M | 440ms | 74 | $7,987/mo | $3,910/mo |
| **Grok 4.7** | `x-ai/grok-4.7` | xAI | frontier | $2.00 | $6.00 | $0.50 | 500K | 310ms | 96 | $42,480/mo | $28,305/mo |
| **Qwen3.8 Omni Flash** | `qwen/qwen3.8-omni-flash` | Alibaba Qwen | open-weight | $0.15 | $0.47 | $0.016 | 1M | 210ms | 155 | $3,244/mo | $1,977/mo |
| **Fugu Ultra v2** | `sakana/fugu-ultra-v2` | Sakana AI | frontier | $5.00 | $30.00 | $0.50 | 1M | 310ms | 96 | $149,400/mo | $106,875/mo |
| **Fugu Max** | `sakana/fugu-max` | Sakana AI | frontier | $2.00 | $6.00 | $0.25 | 1M | 310ms | 96 | $42,480/mo | $25,943/mo |
| **DeepSeek V4.1 Flash** | `deepseek/deepseek-v4.1-flash` | DeepSeek | open-weight | $0.30 | $1.20 | $0.0060 | 1.0M | 210ms | 155 | $7,236/mo | $4,458/mo |
| **GPT-6 Astra Pro** | `openai/gpt-6-astra-pro` | OpenAI | open-weight | $10.00 | $50.00 | $1.00 | 1.1M | 440ms | 74 | $270,000/mo | $184,950/mo |
| **Gemini 3.8 Flash** | `google/gemini-3.8-flash` | Google | fast | $0.825 | $4.13 | $0.083 | 1.0M | 210ms | 155 | $22,275/mo | $15,258/mo |
| **Qwen3.8 Flash** | `qwen/qwen3.8-flash` | Alibaba Qwen | open-weight | $0.15 | $0.47 | $0.016 | 1M | 210ms | 155 | $3,244/mo | $1,977/mo |
| **Seed 2.1 Turbo** | `bytedance-seed/seed-2-1-turbo` | ByteDance Seed | frontier | $0.50 | $2.50 | $0.25 | 262K | 310ms | 96 | $13,500/mo | $11,138/mo |
| **Qwen3.8 2.4T A95B** | `qwen/qwen3.8-2.4t-a95b` | Alibaba Qwen | open-weight | $2.00 | $6.00 | $0.25 | 1.0M | 310ms | 96 | $42,480/mo | $25,943/mo |
| **DeepSeek V4 Pro 0813** | `deepseek/deepseek-v4-pro-0813` | DeepSeek | open-weight | $0.111 | $4.44 | $0.097 | 1.0M | 440ms | 74 | $14,186/mo | $14,054/mo |
| **Claude Opus 5** | `anthropic/claude-opus-5` | Anthropic | frontier | $5.00 | $25.00 | $0.50 | 1M | 310ms | 96 | $135,000/mo | $92,475/mo |
| **Inkling** | `thinkingmachines/inkling` | Thinking Machines | reasoning | $1.00 | $4.05 | $0.17 | 524K | 440ms | 74 | $24,264/mo | $16,421/mo |
| **Claude Sonnet 5** | `anthropic/claude-sonnet-5` | Anthropic | frontier | $2.00 | $10.00 | $0.20 | 1M | 310ms | 96 | $54,000/mo | $36,990/mo |
| **GPT-5.5** | `openai/gpt-5.5` | OpenAI | open-weight | $5.00 | $30.00 | $0.50 | 1.1M | 310ms | 96 | $149,400/mo | $106,875/mo |
| **GPT-5.4** | `openai/gpt-5.4` | OpenAI | open-weight | $2.50 | $15.00 | $0.25 | 1.1M | 310ms | 96 | $74,700/mo | $53,438/mo |
| **Sonar Pro Search** | `perplexity/sonar-pro-search` | Perplexity | reasoning | $3.00 | $15.00 | $1.50 | 200K | 440ms | 74 | $81,000/mo | $66,825/mo |
| **Claude Sonnet 4.5** | `anthropic/claude-sonnet-4.5` | Anthropic | frontier | $3.00 | $15.00 | $0.30 | 1M | 310ms | 96 | $81,000/mo | $55,485/mo |
| **Hermes 4 405B** | `nousresearch/hermes-4-405b` | Nous Research | open-weight | $1.00 | $3.00 | $0.20 | 131K | 310ms | 96 | $21,240/mo | $13,680/mo |
| **GPT-5** | `openai/gpt-5` | OpenAI | open-weight | $1.25 | $10.00 | $0.125 | 400K | 310ms | 96 | $44,550/mo | $33,919/mo |
| **GPT-5 Mini** | `openai/gpt-5-mini` | OpenAI | open-weight | $0.25 | $2.00 | $0.025 | 400K | 210ms | 155 | $8,910/mo | $6,784/mo |
| **Sonar Deep Research** | `perplexity/sonar-deep-research` | Perplexity | frontier | $2.00 | $8.00 | $1.00 | 128K | 310ms | 96 | $48,240/mo | $38,790/mo |

## 2. 2026 Dedicated Cloud GPU On-Demand Pricing Reference (vLLM / SGLang Break-Even)

| GPU Accelerator | VRAM | Typical vLLM Model Fit (4-bit / FP8) | Lowest Verified Cloud Rate ($/hr) | Serverless API Break-Even Volume |
| :--- | ---: | :--- | ---: | :--- |
| **NVIDIA RTX 4090** | 24 GB | QwQ-32B, DeepSeek V4.1 Flash, Mistral Small 24B (4-bit) | $0.34 – $0.69 / hr | >8M tokens/day |
| **NVIDIA L40S** | 48 GB | 32B–70B 4-bit AWQ, multimodal vision-language | $0.79 – $1.35 / hr | >15M tokens/day |
| **NVIDIA H100 SXM / PCIe** | 80 GB | Llama 4 Scout, Mistral Large 4, gpt-oss-120b, 70B FP8 | $1.49 – $2.99 / hr | >42M tokens/day |
| **NVIDIA H200 SXM** | 141 GB | Large 120B–235B MoE + long-context FP8 KV-cache | $2.45 – $3.89 / hr | >55M tokens/day |
| **NVIDIA Blackwell B200** | 192 GB | Sage S2.1 Flash (320B/18B MoE), Qwen4 Max, GLM 5.3 | $3.79 – $6.20 / hr | >75M tokens/day |
| **NVIDIA Blackwell Ultra B300** | 288 GB | Full DeepSeek R1/V4 (671B FP4) on 2x B300, Kimi K3 | $12.50 – $17.02 / hr | >110M tokens/day |

## 3. Multi-Turn Agent Context Bloat Formula (Why 20-Turn Coding Loops Cost 6x–11x Single-Turn Estimates)

For an autonomous coding or tool-use agent running `N` turns with system prompt `S`, per-turn user/tool input `U`, and per-turn output `O`:

- `Effective Avg Input Tokens / Request = S + U * ((N + 1) / 2) + O * ((N - 1) / 2)`
- **Mitigation Stack:** (1) Enable sticky prefix KV-caching (`75%–90%` input discount), (2) pin `provider.order` on OpenRouter so stateless hops do not evict KV-cache, and (3) route `80%` of routine repo search/tool turns to fast models while reserving `20%` of planning turns for frontier models.

## 4. Query TokenMargin Programmatically via MCP (`https://tokenmargin.dev/mcp`)

Add TokenMargin directly to **Claude Code**, **Cursor**, **Windsurf**, or any MCP client (zero API key required):

```json
{
  "mcpServers": {
    "tokenmargin": {
      "type": "http",
      "url": "https://tokenmargin.dev/mcp"
    }
  }
}
```

Available MCP tools:
- `get_llm_pricing`: Lookup live input, output, cached read `$/1M`, context window, TTFT, and throughput by model name, provider, or tier.
- `compare_models`: Compare two LLM APIs head-to-head across monthly DAU volume and prompt cache hit rates.
- `calculate_agent_loop_cost`: Compute exact monthly spend and multi-turn context bloat for `1–20` turn agent workloads.
