Llama 3.1 8B Instant
Groq
$0.08
/month
RealtimeLowest cost128K
Input /1M
$0.05
Output /1M
$0.08
Best for: Ultra-fast simple tasks at 500+ tok/s — latency-sensitive classification and routing
Data reviewed 2026-07-17
AI API Pricing Calculator 2026
Estimate monthly API spend by workload: coding agents, research, RAG support bots, realtime apps, and low-cost automation across OpenAI, Anthropic, Google, xAI, DeepSeek, Mistral, and more.
Cheapest overall
DeepSeek V4 Flash
$0.14 / $0.28 per MTok
Best value general
Gemini 2.5 Flash
$0.30 / $2.50 per MTok
Best for coding
Mistral Codestral
$0.30 / $0.90 per MTok
Most powerful
Claude Opus 4.8
$5.00 / $25.00 per MTok
📋 Not sure about token counts? Pick one or more use cases:
🌍 Your country:No Sales Tax (varies)
ℹ️ No federal VAT. State sales tax on digital services varies (0–10%). Most AI providers do n…
⚠️ Tax rates and exchange rates are approximate and may vary. Actual charges depend on your payment method, billing country, and current exchange rates. Prices shown are estimates only — verify with your provider before subscribing.
💰 Monthly Cost Calculator
~750 words ≈ 1K tokens
~375 words ≈ 500 tokens
= 900/month
Your usage estimate
0.90Minput tokens/mo
0.45Moutput tokens/mo
900requests/mo
⚠️ This estimate excludes:
Taxes, tool/function call costs, web search fees, image/audio input costs, cached write costs, long-context surcharges, and minimum request unit rounding.
Prompt caching can reduce input costs up to 90% for repeated prompts. Actual costs may be 20–50% higher depending on tool/search/image usage.
What this usage means
Quick read before you scan every model row. Costs assume standard pricing only. Discounts (batch, cached input) not applied unless toggled above.
API likely cheaper
Your cheapest API option is about $0.08/mo, below a typical $20 subscription.
Moderate workload
900 requests/month using 1.35M total tokens.
Best price in view
Llama 3.1 8B Instant (Groq) at $0.08/mo.
Workload:
Show every token-priced model in the calculator.
Recommended APIs for this workload
These cards use the selected workload and provider filters, then rank models by monthly cost, quality fit, and production fit.
Cheapest
Lowest bill
Llama 3.1 8B Instant
Groq · $0.08/mo
Start here when the workload is repetitive, simple, or high volume.
Quality
Best output quality
Gemini 3.1 Pro Preview
Google (Gemini API) · $7/mo
Use this when accuracy, reasoning, writing quality, or hard coding tasks matter more than raw price.
Production
Best app backend
Gemini 2.5 Flash-Lite
Google (Gemini API) · $0.27/mo
A practical candidate for RAG, routing, support chat, agents, or cached/batch workloads.
| Model | Provider | Use case fit | Est. Monthly Cost | vs $20 Sub | Input / 1M | Output / 1M | Cached / 1M | Context | Best for |
|---|---|---|---|---|---|---|---|---|---|
Llama 3.1 8B Instant 20M input tokens per dollar. Fastest inference available for this model size. | Groq | RealtimeLowest cost | $0.08 /month | No sub | $0.05 /1M | $0.08 /1M | — | 128K | Ultra-fast simple tasks at 500+ tok/s — latency-sensitive classification and routing |
Command R7B Cheapest flagship-quality API as of June 2026. 4x cheaper than GPT-5.4 Nano on i… | Cohere | RAGLowest costRealtime | $0.10 /month | No sub | $0.0375 /1M | $0.15 /1M | — | 128K | High-volume RAG, classification, routing — cheapest first-party production API |
Llama 4 Scout (DeepInfra) DeepInfra hosted meta-llama/Llama-4-Scout-17B-16E-Instruct. | Meta Llama (hosted via DeepInfra) | Lowest costRealtimeResearch | $0.23 /month | No sub | $0.1 /1M | $0.3 /1M | — | 327,680 | Efficient MoE model — fast and cheap for classification, chat, light coding tasks |
Llama 3.3 70B Turbo (DeepInfra) Current DeepInfra Turbo endpoint; replaces the deprecated non-Turbo endpoint pre… | Meta Llama (hosted via DeepInfra) | Lowest costRealtimeCoding | $0.23 /month | No sub | $0.1 /1M | $0.32 /1M | — | 128K | Cheapest hosted 70B model — best cost/quality for general tasks |
DeepSeek V4 Flash One of the cheapest frontier-class APIs available. Strong on coding benchmarks. | DeepSeek | Lowest costCodingRealtime | $0.25 /month | No sub | $0.14 /1M | $0.28 /1M | $0.0028 cached | 1M | Budget coding tasks, high-volume text generation, agentic pipelines where cost matters most |
Gemini 2.5 Flash-Lite Cheapest model in Google lineup. Flat pricing on all context lengths. | Google (Gemini API) | Lowest costRealtimeRAG | $0.27 /month | API cheaper save $20/mo | $0.1 /1M | $0.4 /1M | $0.01 cached | 1M | Ultra-high-volume simple tasks: classification, routing, lightweight extraction |
Mistral Small 4 Strong price-performance for EU-based applications. | Mistral AI | Lowest costRealtime | $0.41 /month | API cheaper save $20/mo | $0.15 /1M | $0.6 /1M | — | 256K | EU deployments, budget general tasks — French company with EU data residency |
Command R Native grounding and citation generation. Strong for retrieval-heavy workloads. | Cohere | RAGLowest costResearch | $0.41 /month | No sub | $0.15 /1M | $0.6 /1M | — | 128K | RAG pipelines, cost-efficient enterprise chat with native grounding and tool use |
GPT-OSS 120B Current GroqCloud replacement for the retired Kimi K2 endpoint. | Groq | RealtimeLowest cost | $0.41 /month | No sub | $0.15 /1M | $0.6 /1M | — | 128K | Fast open-weight reasoning and agentic workloads on GroqCloud |
Llama 4 Maverick FP8 (DeepInfra) DeepInfra hosted meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8. | Meta Llama (hosted via DeepInfra) | CodingLowest costResearch | $0.54 /month | No sub | $0.2 /1M | $0.8 /1M | — | 1M | Best open-weight model for coding and reasoning — strong competitor to GPT-5.4 |
Codestral Purpose-built coding model. Very competitive pricing vs OpenAI/Anthropic for cod… | Mistral AI | CodingLowest cost | $0.68 /month | API cheaper save $19/mo | $0.3 /1M | $0.9 /1M | — | 128K | Code completion, FIM (fill-in-the-middle), IDE autocomplete, coding-specific tasks — purpose-built for code |
GPT-5.4 Nano Cheapest OpenAI model for production. Official short-context standard pricing. | OpenAI | Lowest costRealtimeRAG | $0.74 /month | API cheaper save $19/mo | $0.2 /1M | $1.25 /1M | $0.02 cached | 400K | Ultra-high-volume simple tasks: classification, intent detection, short-form extraction |
DeepSeek V4 Pro Current official price from DeepSeek API docs. Legacy deepseek-chat and deepseek… | DeepSeek | CodingLowest cost | $0.78 /month | No sub | $0.435 /1M | $0.87 /1M | $0.003625 cached | 1M | Complex coding and reasoning at significantly lower cost than OpenAI/Anthropic equivalents |
Llama 3.3 70B Versatile ~250 tok/s. Undercuts GPT-4o mini on output cost while offering flagship-class c… | Groq | RealtimeLowest costCoding | $0.89 /month | No sub | $0.59 /1M | $0.79 /1M | — | 128K | Flagship-quality at low cost — coding, QA, summarization on Groq LPU infrastructure |
Gemini 3.1 Flash-Lite Current cost-efficient Gemini 3 model. Batch/Flex standard text pricing is $0.12… | Google (Gemini API) | Lowest costRealtimeRAG | $0.90 /month | API cheaper save $19/mo | $0.25 /1M | $1.5 /1M | $0.025 cached | 1M | High-volume agentic tasks, translation, lightweight data processing |
Mistral Large 3 Flagship Mistral model. Good alternative to GPT-4o for EU deployments. | Mistral AI | QualityResearchLowest cost | $1 /month | API cheaper save $19/mo | $0.5 /1M | $1.5 /1M | — | 256K | Complex reasoning, EU compliance-sensitive tasks, multilingual European deployments |
Sonar Search included at no extra charge. ~$5-12 per 1,000 requests additional based o… | Perplexity (Sonar API) | ResearchRAG | $1 /month | No sub | $1 /1M | $1 /1M | — | 127K | Web-grounded Q&A with citations — cheapest model with built-in real-time search |
Gemini 2.5 Flash Best price-performance in current Gemini lineup. Free tier available. | Google (Gemini API) | RealtimeRAGLowest cost | $1 /month | API cheaper save $19/mo | $0.3 /1M | $2.5 /1M | $0.03 cached | 1M | High-volume general tasks: summarization, extraction, coding assistance, rapid prototyping |
grok-build-0.1 Early-access coding model listed on xAI API pricing. | xAI (Grok API) | CodingRealtime | $2 /month | API cheaper save $18/mo | $1 /1M | $2 /1M | $0.2 cached | 256K | Agentic coding and build tasks on xAI infrastructure |
Kimi K2.5 Official Kimi API price: cache hit $0.10, cache miss $0.60, output $3.00 per MTo… | Moonshot AI (Kimi) | CodingLowest costResearch | $2 /month | API cheaper save $18/mo | $0.6 /1M | $3 /1M | $0.1 cached | 256000 | Budget coding and multimodal tasks — cheapest vision-capable model in this comparison |
Grok 4.3 Current xAI flagship text model. | xAI (Grok API) | QualityResearchRealtime | $2 /month | API cheaper save $18/mo | $1.25 /1M | $2.5 /1M | $0.2 cached | 1M | General reasoning, tool calling, long-context applications |
Kimi K2.6 Official Kimi API price: cache hit $0.16, cache miss $0.95, output $4.00 per MTo… | Moonshot AI (Kimi) | CodingResearch | $3 /month | API cheaper save $17/mo | $0.95 /1M | $4 /1M | $0.16 cached | 256000 | Coding and agentic tasks at 4× lower cost than Claude Sonnet — strong coding benchmark scores |
GPT-5.4 Mini Mid-range GPT-5.4 family model. Official short-context standard pricing. | OpenAI | CodingResearch | $3 /month | API cheaper save $17/mo | $0.75 /1M | $4.5 /1M | $0.075 cached | 400K | Lower-cost coding, long-form summarization, and medium-complexity agent steps |
Claude Haiku 4.5 Cheapest current-gen Claude. Replaces deprecated Haiku 3 ($0.25/$1.25). | Anthropic | RealtimeRAG | $3 /month | API cheaper save $17/mo | $1 /1M | $5 /1M | $0.1 cached | 200K | Classification, routing, extraction, summarization, high-volume workloads |
GPT-5.6 Luna | OpenAI | RealtimeCoding | $4 /month | API cheaper save $16/mo | $1 /1M | $6 /1M | $0.1 cached | 1.05M | Fastest and cheapest GPT-5.6 — good for high-volume, latency-sensitive tasks |
Moonshot V1 (128K) Legacy model. Official price is $2 input/$5 output per MTok; platform sunset exp… | Moonshot AI (Kimi) | Research | $4 /month | API cheaper save $16/mo | $2 /1M | $5 /1M | — | 128000 | Legacy model — superseded by K2.5 and K2.6. Listed for reference only. |
Gemini 2.5 Pro Best value for complex tasks in Gemini lineup. 2x surcharge beyond 200K. | Google (Gemini API) | ResearchCodingQuality | $6 /month | API cheaper save $14/mo | $1.25 /1M | $10 /1M | $0.125 cached | 1M | Complex reasoning, coding, long-document analysis — strong at 1M context tasks |
Claude Sonnet 5 Introductory pricing through 2026-08-31; standard pricing is $3/$15 per MTok the… | Anthropic | CodingResearch | $6 /month | API cheaper save $14/mo | $2 /1M | $10 /1M | $0.2 cached | 1M | High-performance coding, agents, and general production workloads during introductory pricing |
Command R+ Same price as GPT-5.4 on input, 33% cheaper on output. Purpose-built for RAG bea… | Cohere | RAGResearchQuality | $7 /month | No sub | $2.5 /1M | $10 /1M | — | 128K | Complex agentic RAG, enterprise document intelligence, multi-step grounded reasoning |
Gemini 3.1 Pro Preview Preview model. Standard price for prompts up to 200K tokens; prompts above 200K … | Google (Gemini API) | QualityResearchCoding | $7 /month | API cheaper save $13/mo | $2 /1M | $12 /1M | $0.2 cached | 1M | High-quality multimodal reasoning, agentic workflows, vibe-coding |
GPT-5.4 Strong balance of capability and cost. Official short-context standard pricing; … | OpenAI | CodingResearch | $9 /month | API cheaper save $11/mo | $2.5 /1M | $15 /1M | $0.25 cached | 1.05M | Coding (debugging, code gen, refactoring), content generation, analysis — solid all-rounder |
GPT-5.6 Terra | OpenAI | CodingResearch | $9 /month | API cheaper save $11/mo | $2.5 /1M | $15 /1M | $0.25 cached | 1.05M | Balanced quality and cost — good daily driver for most tasks |
Claude Sonnet 4.6 Legacy/current comparison point. Sonnet 5 is cheaper during introductory pricing… | Anthropic | CodingResearchRAG | $9 /month | API cheaper save $11/mo | $3 /1M | $15 /1M | $0.3 cached | 1M | General coding, analysis, writing, RAG pipelines, agentic tasks |
Sonar Pro $6-14 per 1,000 requests additional. Use when citation accuracy and source depth… | Perplexity (Sonar API) | ResearchRAG | $9 /month | No sub | $3 /1M | $15 /1M | — | 127K | Deep research with multi-source citations — complex questions requiring current web data |
Claude Opus 4.8 Current Opus model. Fast Mode is 2x standard pricing. | Anthropic | QualityCodingResearch | $16 /month | API cheaper save $4/mo | $5 /1M | $25 /1M | $0.5 cached | 1M | Complex reasoning, agentic workflows, code review, long-context analysis |
GPT-5.5 Flagship model. Official short-context standard pricing; long-context and Priori… | OpenAI | QualityCodingResearch | $18 /month | Break-even | $5 /1M | $30 /1M | $0.5 cached | 1.05M | Complex multi-step workflows, frontier writing quality, tool-use intensive agents |
GPT-5.6 Sol | OpenAI | QualityCodingResearch | $18 /month | Break-even | $5 /1M | $30 /1M | $0.5 cached | 1.05M | Flagship — highest quality for complex reasoning, coding, and research |
Claude Fable 5 Top-tier Claude model listed in official pricing. More expensive than Opus 4.8. | Anthropic | QualityResearchCoding | $32 /month | Sub cheaper $12/mo over | $10 /1M | $50 /1M | $1 cached | 1M | Frontier reasoning, long-context agentic workflows, high-stakes analysis |
Claude Mythos 5 Limited availability. Official Claude Platform pricing is $10/$50 per MTok. | Anthropic | QualityResearch | $32 /month | Sub cheaper $12/mo over | $10 /1M | $50 /1M | $1 cached | 1M | Advanced reasoning and long-context work where limited availability is acceptable |
GPT-5.5 Pro Premium model. Official pricing is $30/$180 per MTok; no cached-input discount. | OpenAI | QualityResearch | $108 /month | Sub cheaper $88/mo over | $30 /1M | $180 /1M | — | 1.05M | Most demanding frontier tasks where quality matters more than cost |
Prices shown are standard per-token rates. Batch, cached, priority, and long-context rates vary. Verify current pricing at each provider's official API documentation.
Provider notes & discounts
Anthropic
Official pricing ↗Output tokens are typically priced 5x input on Claude models. Batch API: 50% off. Prompt caching read is 90% off cached input. Sonnet 5 is listed with introductory $2/$10 pricing through 2026-08-31; standard pricing is $3/$15 thereafter.
✓ batch: 50% off (24hr async processing)✓ prompt cache: 90% off cached input tokens✓ long context surcharge: 2x input price beyond 200K tokens on some models
OpenAI
Official pricing ↗Standard short-context pricing shown for flagship models. OpenAI lists separate long-context and Priority pricing for GPT-5.5/GPT-5.4 families. Batch and Flex pricing may differ by endpoint; verify current rates before production use.
✓ batch: 50% off (24hr async)✓ prompt cache: 90% off cached input (GPT-5.5 and 5.4 families)✓ priority: Premium pricing for guaranteed faster processing; see official table
Google (Gemini API)
Official pricing ↗Free tier available on many Gemini models. Batch and Flex are often discounted. Context caching is priced separately by model. Gemini 3 pricing includes 5,000 free grounded prompts per month shared across Gemini 3 models.
✓ batch: 50% off (24hr async)✓ context cache: 90% off cached input✓ free tier: Free tier available on many Gemini models✓ long context surcharge: 2x input/output beyond 200K tokens on Gemini 2.5 Pro standard pricing
xAI (Grok API)
Official pricing ↗Current xAI API text pricing is focused on Grok 4.3 and grok-build-0.1. Image, video, and voice APIs use per-image, per-second, or per-hour pricing.
✓ new user credits: $25 free✓ data sharing: $150/month additional via data sharing program
DeepSeek
Official pricing ↗DeepSeek V4 Flash and V4 Pro both support 1M context and 384K max output. Privacy considerations: Chinese company. Web chat is free.
Mistral AI
Official pricing ↗EU-based provider, strong GDPR compliance. Open-source models available. Good for EU privacy-conscious deployments.
Cohere
Official pricing ↗Purpose-built for enterprise RAG (Retrieval-Augmented Generation). Command R7B is cheapest first-party production API in 2026. Complete RAG stack: Embed v3 ($0.10/M), Rerank 3.5 ($2/1K queries), Command generation. Free trial key available.
✓ free trial: Rate-limited trial key, no credit card required✓ enterprise: Custom pricing on AWS, Azure, Oracle, self-hosted
verified 2026-07-17
GroqCloud hosted pricing. Llama 3.1 8B and Llama 3.3 70B are scheduled for developer-tier shutdown on 2026-08-16; GPT-OSS 120B is the recommended current replacement.
✓ free tier: All models free with rate limits✓ batch: 50% off✓ prompt cache: 50% off cached tokens
Perplexity (Sonar API)
Official pricing ↗Sonar prices include token charges plus a separate per-request search-context fee. Low/medium/high context fees differ by model; token-only estimates understate total cost.
✓ vs gpt search: Typically 3-5x cheaper than GPT-5.4 + web search for research workloads
Meta Llama (hosted via DeepInfra)
Official pricing ↗Meta does not sell a direct Llama API. These are current DeepInfra hosted endpoints and prices, verified from DeepInfra's first-party model-list API.
✓ self hosted: Free if self-hosted — only pay for GPU infrastructure✓ deepinfra: Current listed endpoints: Scout $0.10/$0.30, Maverick FP8 $0.20/$0.80, Llama 3.3 70B Turbo $0.10/$0.32
FLUX API (Black Forest Labs)
Official pricing ↗Per-image pricing, not per-token. 1 credit = $0.01. Prices vary by model, operation, and output resolution.
✓ aggregators: FAL.ai and Replicate often 30-50% cheaper than BFL direct for older versions✓ self hosted: FLUX.2 [klein] 4B free with ~13GB VRAM
Moonshot AI (Kimi)
Official pricing ↗Prices exclude applicable taxes. Kimi K2.6 and K2.5 prices were extracted from the official page source because the rendered table parser omitted the numeric cells. Moonshot V1 platform sunset is expected on 2026-08-31.
API vs subscription: which is cheaper for you?
At ~2,000–2,200 interactions/month, Claude Sonnet API and Claude Pro subscription cost roughly the same. Below that, API wins. Above it, the flat $20/month Pro subscription is cheaper. Read our full breakdown:
API vs Subscription: When does pay-per-token save money? →All 12 providers and 42 listed models rechecked against first-party pricing pages on 2026-07-17. Corrected OpenAI, Google, Anthropic, Mistral, Meta-hosted, Groq, and FLUX records; model-level verification metadata is complete. Workload tags were added on 2026-07-17 as a first-pass editorial classification for API pricing UX. Cost calculator uses standard pricing; batch and caching discounts apply separately. Actual costs may vary based on model routing, context length, and feature usage.