Skip to content

LLM API Pricing & AI Token Cost Calculator

Calculate, compare, and forecast API token expenditures across 50+ frontier models from OpenAI, Anthropic, DeepSeek, Meta, Google, and NVIDIA. Features live pricing for GPT-6 Astra, Claude Sonnet 5, Llama 4 Maverick, DeepSeek V4.1, and Nemotron 3.

API Token Budgeting Experience

Quick & Simple: Popular workload presets, streamlined model pickers, and immediate monthly run rates.

Experience Mode:Quick & SimpleSynced with Site (Standard)

Streamlined view with essential inputs, clear verdicts, and zero cognitive overload.

Standard Architecture Presets
🧠 Reasoning tokens constitute 32.4% of total output tokens.
Per-Call Cost
$0.05
25,000 in + 3,700 out billed
Monthly Run Rate
$133
2,500 calls/month volume
Annual Run Rate (12-Mo)
$1,598
Unadjusted 1-year extrapolation

Production Budget & ROI Analysis

High-Efficiency Choice: Claude Sonnet 5 operates at just $2.00/1M input, making it ideal for high-frequency agent loops, batch document ingestion, and continuous enterprise background workers.

Custom Token Sizing, Reasoning Tokens & Caching Controls▾ Expand Custom Tuning

Workload Token Sizing

25,000 tokens
100200K500K1M
2,500 tokens
508K32K64K
2,500 calls/month
100 calls50K250K1M calls

🖼️ Vision Tokenizer

Reasoning / Thinking Tokens (CoT)
Model Supports CoT

Models generate internal reasoning tokens billed at Output rate ($10.00/1M).

Charged Output:2500 (visible) + 1200 (thinking) = 3,700 tokens
Architectural Cost Reductions
Up to 90% savings
~80% read discount
Static Prefix Ratio:
Caching Monthly Savings:$84.37/mo

Claude Sonnet 5 Rate CardAnthropic

Input Rate$2.00 / 1M
Output Rate$10.00 / 1M
Context Window1,000,000
Blended Cost (30/70)$7.60 / 1M

Cross-Model Cost Comparison for Identical Workload

Click any row to select model
Model & ProviderContextCost per CallMonthly TotalDifference vs Claude Sonnet 5Switch
Claude Sonnet 5Current
Anthropic
1,000,000$0.05$133Baseline
AAion 3.5
Aion-labs
256,000$0.09$225+69% more
AAion 3.5 Mini
Aion-labs
256,000$0.02$52.5061% cheaper
Qwen 3.8 Flash
Alibaba
1,000,000$0.0025$6.1695% cheaper
Qwen 3.8 Max
Alibaba
1,000,000$0.04$98.4726% cheaper
Qwen3.8 Max Prime
Alibaba
1,000,000$0.06$161+21% more
Qwen3.8 Omni Flash
Alibaba
1,000,000$0.0025$6.1695% cheaper
Qwen4 Max
Alibaba
1,000,000$0.02$60.3555% cheaper
Claude Fable 5.1
Anthropic
1,000,000$0.27$666+400% more
Claude Fable 5.2
Anthropic
1,000,000$0.17$413+210% more
Claude Haiku 5.5
Anthropic
1,000,000$0.0021$5.1696% cheaper
Claude Opus 5.5
Anthropic
1,000,000$0.19$486+265% more
Claude Sonnet 5.5
Anthropic
1,000,000$0.04$10323% cheaper
ERNIE-Light-DT-1.0
Baidu
1,000,000$0.04$95.2528% cheaper
Command A+
Cohere
192,000$0.01$28.1379% cheaper
DeepSeek V4.1 Flash
DeepSeek
1,000,000$0.0031$7.7194% cheaper
DeepSeek V4.1 Flash 1017
DeepSeek
1,000,000$0.0031$7.8594% cheaper
EGlimpse
Elorian
256,000$0.02$56.2558% cheaper
Ember-1
Fireworks
1,000,000$0.11$281+111% more
Gemini 3.1 Pro
Google
1,000,000$0.05$1247% cheaper
Gemini 3.8 Flash
Google
2,000,000$0.0060$14.9289% cheaper
Gemini 4 Argon
Google
1,000,000$0.07$188+41% more
Gemini 4 Carbon
Google
1,000,000$0.07$188+41% more
Gemini 4 Hydrogen
Google
1,000,000$0.06$141+6% more
Gemini Omni 1.1 Flash
Google
128,000$0.06$150+13% more
Gemini Omni 1.2 Flash
Google
128,000$0.06$150+13% more
Nano Banana 2.1
Google
66,000$0.06$141+6% more
IMercury 2.5
Inception
256,000$0.0014$3.4497% cheaper
IMercury 3
Inception
1,000,000$0.0034$8.4494% cheaper
ILing 3.0 Flash VL
inclusionAI
256,000$0.00068$1.7099% cheaper
ILing 3.1 Flash
inclusionAI
256,000$0.0024$6.0995% cheaper
ISchematron V2 Small
Inference-net
128,000$0.0018$4.5697% cheaper
ISchematron V2 Turbo
Inference-net
128,000$0.0011$2.8198% cheaper
KMi:dm K 3.0 Pro
Kt
256,000$0.0070$17.5087% cheaper
LLFM2.5-24B-A2B
Liquid
128,000$0.0030$7.5094% cheaper
LLFM3-1.7B
Liquid
256,000$0.0025$6.2595% cheaper
Llama 4 Maverick (401B MoE)
Meta
1,000,000$0.0043$10.7492% cheaper
Llama 4 Scout (70B)
Meta
512,000$0.0014$3.5697% cheaper
Muse Ember
Meta
1,000,000$0.0028$7.0395% cheaper
Muse Spark 1.3
Meta
1,000,000$0.03$75.3943% cheaper
Muse Spark Open
Meta
1,000,000$0.03$75.3943% cheaper
Muse Wonder
Meta
1,000,000$0.11$281+111% more
MiniMax M3.1 Flash
Minimax
1,000,000$0.0052$13.1290% cheaper
Mistral Large 4
Mistral
1,000,000$0.01$26.8780% cheaper
MMoondream 3 Preview
Moondream
32,000$0.01$34.3874% cheaper
MMoondream 3.1
Moondream
32,000$0.01$34.3874% cheaper
Kimi K3
Moonshot
200,000$0.06$161+21% more
Kimi K3.1 Pro
Moonshotai
1,000,000$0.06$141+6% more
MMotif 3
Motif-technologies
256,000$0.01$26.2580% cheaper
NNex-N2.6
Nex-agi
256,000$0.0040$10.0092% cheaper
Nemotron 3 Nano (30B)
NVIDIA
128,000$0.0021$5.2596% cheaper
Nemotron 3 Super (120B)
NVIDIA
256,000$0.0025$6.3595% cheaper
Nemotron 3 Ultra (550B)
NVIDIA
256,000$0.02$52.5061% cheaper
Nemotron 3.5 Lightning
NVIDIA
256,000$0.0014$3.4497% cheaper
Nemotron 3.5 Super VL
NVIDIA
1,000,000$0.0035$8.7593% cheaper
Switchyard
NVIDIA
1,000,000$-27500.000000$-68750000.00000051643292% cheaper
GPT-5.6 Sol
OpenAI
200,000$0.06$141+6% more
GPT-6 Astra
OpenAI
1,000,000$0.34$853+541% more
GPT-6 Luna
OpenAI
1,100,000$0.0028$7.0395% cheaper
GPT-6 Luna Pro
OpenAI
1,000,000$0.0028$7.0395% cheaper
GPT-6 Sol
OpenAI
1,100,000$0.07$171+28% more
GPT-6 Sol Pro
OpenAI
1,000,000$0.06$141+6% more
GPT-6.1 Sol
OpenAI
1,000,000$0.06$141+6% more
GPT-6.1 Sol Pro
OpenAI
1,000,000$0.06$141+6% more
GPT-6.1 Terra
OpenAI
1,000,000$0.03$70.3147% cheaper
GPT-6.1 Terra Pro
OpenAI
1,000,000$0.03$70.3147% cheaper
PParticle-1
Particle-ai
1,000,000$0.0025$6.2595% cheaper
PPerceptron Mk1.5
Perceptron
37,000$0.0075$18.7586% cheaper
PTernary Bonsai 2 27B
Prism-ml
256,000$0.0031$7.8194% cheaper
PTernary Bonsai 2 9B
Prism-ml
256,000$0.0023$5.8196% cheaper
RBeam 1.0 501B A23B
Reflection
1,000,000$0.03$81.2539% cheaper
RRipple 65B Preview
Ricursiveai
1,000,000$0.0065$16.2588% cheaper
SFugu Max
Sakana AI
1,000,000$0.07$163+22% more
SFugu Ultra v2
Sakana AI
1,000,000$0.17$414+211% more
SExtract
Scaledown
1,000,000$0.0010$2.5098% cheaper
SSummarize
Scaledown
1,000,000$0.0010$2.5098% cheaper
Step 5 Preview
Stepfun
1,000,000$0.03$79.3840% cheaper
Hy4
Tencent
1,000,000$0.01$33.8875% cheaper
TInkling 2 Small
Thinkingmachines
1,000,000$0.01$25.0081% cheaper
TJev Router
Typesafe
1,000,000$-27500.000000$-68750000.00000051643292% cheaper
UPareto
Unbiased
256,000$0.08$203+53% more
UPareto 26.10
Unbiased
1,000,000$0.03$70.0047% cheaper
UPareto 26.10 Preview
Unbiased
1,000,000$0.03$70.0047% cheaper
UPareto 26.9
Unbiased
256,000$0.08$203+53% more
USolar Mini 4
Upstage
524,000$0.0018$4.3897% cheaper
Grok 4 Vision
xAI
500,000$0.05$1228% cheaper
Grok 4.6
xAI
500,000$0.07$163+22% more
Grok 4.7
xAI
500,000$0.07$181+36% more
XMiMo-V2.6-Flash
Xiaomi
1,000,000$0.0042$10.5092% cheaper
XMiMo-V2.6-Pro
Xiaomi
1,100,000$0.0080$19.9485% cheaper
XMiMo-V2.6-Pro-UltraSpeed
Xiaomi
1,000,000$0.13$326+145% more
ZGLM 5.3 Flash
Z.ai
1$0.00052$1.3099% cheaper
ZGLM 5.3 FlashX
Z.ai
1,000,000$0.0061$15.3388% cheaper
ZGLM 5.3 FlashX Legal
Z.ai
1,000,000$0.0061$15.3388% cheaper
ZGLM 5.3 Prime
Z.ai
1,000,000$0.04$11216% cheaper
Cloud GPU Infrastructure

Need More VRAM? Deploy On-Demand Cloud GPUs

Bypass local memory limits. Spin up an RTX 4090 or A100 80GB in under 60 seconds with per-minute billing.

RunPodMost Popular
$0.29/hrstarting rate

Instant cloud GPUs with 1-click vLLM, Ollama, and PyTorch templates. RTX 4090 to 8x H100.

Best for: Ollama, vLLM & Rapid LLM Serving
Vast.aiLowest Price
$0.18/hrstarting rate

Decentralized GPU marketplace with aggressive per-hour rates for consumer & datacenter cards.

Best for: Budget LoRA Fine-Tuning & Batch Inference
Lambda LabsEnterprise Performance
$0.50/hrstarting rate

High-speed networked enterprise clusters with low latency and dedicated NVMe storage.

Best for: Multi-GPU Training & Production Serving

Rates verified weekly. Transparent disclosure: We may receive an affiliate commission from partner signups at no extra cost to you.

Engineering LLM API Economics: Complete Cost Optimization Guide

Building scalable infrastructure on frontier foundation models requires deterministic unit economic modeling. Without proactive architectural safeguards, compounding context windows, un-cached system prompts, and unmonitored internal reasoning traces can inflate monthly operational expenditures by 300% to 800%.

How are 2026 LLM API token prices structured across input, output, and reasoning tokens?

Token / Billing TierData Unit2026 Rate Range ($/1M Tokens)Technical Behavior & Cost Constraints
Standard Input (Prompt)per 1M tokens$0.02 – $2.50Parallel prefill pass. Scaled by system prompts, conversation history, and RAG chunks.
Standard Output (Completion)per 1M tokens$0.08 – $10.00Autoregressive generation pass. Memory bandwidth bound; priced 3× to 5× higher than input.
Reasoning / Thinking Tokensper 1M tokens$0.42 – $50.00Internal Chain-of-Thought (CoT). Hidden in UI but charged strictly at full Output Token pricing.
Cached Prompt Readper 1M tokens$0.02 – $0.50Re-reads existing prefix KV-cache in memory. Provides 75% to 90% direct price reduction.
Batch API Requestsper 1M tokens50% DiscountAsynchronous, non-realtime batch queue with guaranteed 24-hour turnaround SLA.

How do Prompt Caching and Batch API discounts reduce production LLM expenditures?

Deterministic Production API Cost Formula:

Total Cost = [(Uncached Input × P_in) + (Cached Input × P_cache) + ((Output + Reasoning) × P_out)] × (1 − Batch_Discount)

Example Calculation (Claude Sonnet 5: 1,000 calls with 25K prompt, 80% cached, 2K output, 4K reasoning):

Uncached Input: 5M tokens × $2.00/1M = $10.00

Cached Read Input: 20M tokens × $0.20/1M = $4.00

Output + Reasoning: 6M tokens × $10.00/1M = $60.00

Total Call Expense: $74.00 (vs. $135.00 without caching — 45.2% instant savings)

Model NameProviderContext WindowInput / 1MOutput / 1MCache Read RateProduction Architecture Tier
DeepSeek V4.1 FlashDeepSeek1,000,000$0.14$0.42$0.035 (75% off)Tier 1 — Ultra-Fast Router & Triage
Ling 3.0 Flash VLInclusionAI256,000$0.02$0.06$0.005 (75% off)Tier 1 — High-Volume Event Ingestion
Llama 4 MaverickMeta (Open-Weights)1,000,000$0.188$0.653$0.050 (73% off)Tier 2 — Primary Synthesis Workhorse
Gemini 3.8 FlashGoogle1,000,000$0.75$3.75$0.187 (75% off)Tier 2 — Long-Context Multimodal Agent
Claude Sonnet 5Anthropic1,000,000$2.00$10.00$0.200 (90% off)Tier 3 — Flagship Engineering & Coding
Nemotron 3 Super (120B)NVIDIA256,000$0.08$0.24$0.020 (75% off)Tier 2 — Enterprise On-Prem & Private Cloud
GPT-6 AstraOpenAI1,000,000$10.00$50.00$2.500 (75% off)Tier 3 — Maximum Reasoning & Scientific Proofs

Recommended 3-Tier Production Cascade

Leading engineering teams avoid routing 100% of user traffic directly to expensive Tier 3 flagship models. Instead, they deploy a 3-tier cascaded architecture:

  1. Tier 1 — Ultra-Fast Routers & Triage ($0.02–$0.15/1M): Deploy ultra-efficient models like DeepSeek V4.1 Flash, Ling 3.0 Flash VL, or Mercury 2.5 to parse user intent, run safety guardrails, and execute lightweight SQL or CRM queries.
  2. Tier 2 — Core Workhorse Synthesis ($0.18–$0.75/1M): Handle 80% of standard production generation, chat dialogues, code drafting, and summarization using Llama 4 Maverick, Gemini 3.8 Flash, or Nemotron 3 Super.
  3. Tier 3 — Complex Reasoning Escalation ($2.00–$10.00+/1M): Escalate only multi-file architecture refactoring, complex mathematical validation, or autonomous agent decision loops to Claude Sonnet 5, GPT-6 Astra, or DeepSeek-R1.

Frequently Asked Questions

How are 2026 LLM API token prices structured across input, output, and reasoning tokens?

In 2026, foundation models charge across three distinct tiers: Input Tokens ($0.02 to $10.00/1M) processed in parallel across attention heads; Output Tokens ($0.08 to $50.00/1M) computed sequentially token-by-token; and internal Reasoning / Thinking Tokens (generated by models like GPT-6 Astra, Claude Sonnet 5, and DeepSeek-R1) which are billed at the higher Output Token rate, often doubling or tripling effective per-request costs.

How do Prompt Caching and Batch API discounts reduce production LLM expenditures?

Prompt Caching allows providers (Anthropic, DeepSeek, Google, OpenAI) to persist compiled prompt prefixes in GPU memory across requests, granting 75% to 90% discounts on cache reads ($0.10 to $0.50/1M). Asynchronous Batch APIs provide a guaranteed 50% discount on standard token rates for non-urgent jobs processed within a 24-hour SLA.

What is the cost difference between open-weights hosting and proprietary foundation APIs?

Open-weights models such as Llama 4 Maverick (401B MoE) and Nemotron 3 Super (120B) offer token rates as low as $0.188 input / $0.653 output per 1M tokens through optimized cloud inference providers, compared to $2.00–$10.00 input / $10.00–$50.00 output for proprietary frontier flagships like Claude Fable 5.1 and GPT-6 Astra.

How do you accurately estimate token counts for unstructured developer documents?

On modern byte-pair encoding (BPE) tokenizers like o200k_base, 1 English token corresponds to roughly 4 characters or 0.75 words. A 250-word book page corresponds to 330–350 tokens, a 500-word document consumes ~670 tokens, and 1,000 lines of typical TypeScript code averages 8,000–10,500 tokens.