Kimi K3 in vLLM: faster decode is not an API price cut
The 842 tokens/s headline uses 16 GB300 GPUs and synthetic speculative acceptance. Compare the benchmark limits, current cache-write billing, and self-hosting cost math.
Read the Kimi K3 cost analysis →Compare AI Token & Subscription Prices With Daily-Verified Data
Check AI pricing before the bill lands. We track API token pricing for 243 models across 19 providers, plus monthly AI subscription plans for ChatGPT, Claude, Gemini, Perplexity, Copilot, and more. Current cheapest flagship API model: DeepSeek V4.1 Flash at $0.15 per million input tokens.
October 14 · Amp with OpenAI
Agents beyond code: what ChatGPT covers—and what it doesn't
Astra design, Puck voice and Plaid speed have different billing paths. Compare maintained plan, runner and API rates before scaling.
October 14 · OpenAI and Third Moment Research
10,000 research agents: scale is not a price cut
36–48-hour sessions, Sol-led routing and a faster backtest. Compare live API prices and the evidence needed to estimate a real fleet bill.
October 14 · Physical Intelligence Pi API beta
Robot-model training and inference: access opens, prices remain undisclosed
Managed fine-tuning and hosted inference are in a batch-onboarded beta. No public rate card, free tier or universal robot compatibility is confirmed.
October 11 · Anthropic policy and pricing
Can Claude end abusive chats? The rule is narrower than the headline
The new rule takes effect November 12 and targets sustained, needless abuse. Ordinary pushback and testing are excluded; Claude API rates did not change.
October 9 · Decision-model pricing
Jevman: what does AI Pac-Man reveal about cost per decision?
Six models, 100 games each. Jev's direct API rate is $0.042 per million input tokens; Opper gateway fees and game costs are separate.
October 8 · AI finance cost analysis
AI beats an earnings-consensus baseline—but what does each forecast cost?
Samaya's finance harness tested seven models; no API rate or end-to-end forecast bill was announced.
October 8 · Creative workflow pricing
Pollo AI uses OpenAI to build ad campaigns—what does a finished ad cost?
GPT-5.6 routes work, Astra plans harder scenes, and GPT-Image-2.5 creates images. API rates are unchanged; the video bill is undisclosed.
October 7 · ChatGPT rollout
GPT-6 Intelligent UI reaches ChatGPT paid tiers; Free and Go follow
Sol powers paid Chat, Luna powers Free and Go. No new subscription or API rate was announced; Work and Codex are unchanged.
October 7 · Research and pricing update
OpenAI shares AI mathematics results—but not an API price
The results come from an internal frontier model. Its Pro-equivalent compute estimate is not a customer bill; existing API rates are unchanged.
October 6 · Enterprise AI update
Atlassian expands OpenAI models across Rovo and agents
GPT-6 Astra and GPT-5.6 access is named, but no new public API rate or Rovo price is announced. Separate Rovo credits from direct model-token costs.
Measured · AI Pricing Guru Labs
Accuracy is only half the model-cost story
10 models tie for best at 100% accuracy, while GPT-6 Luna delivers the lowest measured cost per correct answer.
AI API price scale
How leading flagship and frontier AI model prices compare
One flagship plus one generally available frontier model may appear per creator. Limited-access specialist variants and superseded generations are excluded. Each column uses the cheapest active host.
Choose your pricing type
API tokens, monthly plans, or exact cost math?
Compare developer API rates separately from consumer subscription pricing, with calculators linked from both.
API token pricing
Compare input, output, cached, and batch rates per 1M tokens across model providers.
View token prices → For users & teamsSubscription pricing
Compare ChatGPT Plus, Claude Pro, Gemini, Perplexity, Copilot, and higher-tier monthly plans.
View subscriptions →Token cost calculator
Estimate your API bill from token volume, caching, workload presets, and current model rates.
Calculate token cost → · All calculatorsSelf-hosting costs
GPU requirements, rented-rig monthly bills, and the break-even volume against API pricing for open models.
Map self-host costs →Top Models at a Glance
A small sample from the live pricing table. Prices are per 1M tokens.
Cost snapshot
Need exact math?
Here is the short version for 1M input + 1M output tokens. Use the full calculator when you need custom token counts, caching, filters, and model-by-model sorting.
Latest major price move
Claude Mythos 5.1 is now 83% cheaper
Anthropic's published standard input and output rates both fell. For Claude Mythos 5.1, 1M input + 1M output tokens dropped from $60.00 to $10.25. The comparison is generated from consecutive daily snapshots, so the homepage advances when a newer material price cut lands.
Before vs now
Cost of 1M input + 1M output tokens
Source: daily pricing history snapshots. Current rates are rechecked by the pricing pipeline; historical comparisons come from the latest consecutive snapshot change.
Latest pricing intelligence
Latest AI pricing news
A compact feed of fresh launches, billing changes, and cost-saving guides.
Kimi K3 vLLM Speedup — Pricing Impact (Oct 2026)
vLLM reports Kimi K3 decode gains on GB300. See the 842 tokens/s caveats, self-hosting cost math, and current managed API prices.
Read →Google Gemini Self-Improvement — Pricing Impact (Oct 2026)
Sergey Brin says Gemini is approaching recursive self-improvement. Google's safety report adds caveats; see current Argon API access and pricing.
Read →Gemini 4 Argon Pricing & Access: What Changed
Gemini 4 Argon now has a stable paid API endpoint. Compare live prices, December rate changes, context limits, benchmarks, and Labs coverage.
Read →Haiku 5.5 Pricing: The 100k Context Threshold
Haiku 5.5 reprices the whole request above 100,000 prompt tokens. Understand cache accounting and calculated costs before choosing a context budget.
Read →Microsoft MAI-Base-2: Pricing and API Availability
Microsoft details MAI-Base-2 frontier pre-training and a claimed 4.2–12.3× efficiency gain. What it means for API costs, Foundry licensing, and buyers.
Read →OpenAI Amp Agents Beyond Code — Pricing Impact (Oct 2026)
Amp uses OpenAI models for planning, writing, design, and voice. See ChatGPT billing exclusions, Plaid's 6× token cost, and current API prices.
Read →Third Moment's 10,000 OpenAI Agents: Cost and Pricing
Third Moment runs 10,000 agents for 36–48 hours. See live Astra, Sol and Luna prices, the Codex backtest claim, and what the bill leaves unknown.
Read →Physical Intelligence API Beta — Pricing Impact (Oct 2026)
Physical Intelligence opens a robot-model API beta. See what is available, why pricing remains unknown, and what to ask before committing.
Read →Spotatui Spotify Rate Limits: AI DJ Costs (Oct 2026)
Spotatui uses Spotify's internal API and offers an AI DJ. See Premium requirements, subscription quota tradeoffs, and live model costs.
Read →Higgsfield 10-Minute AI Video: Astra Pricing Impact
Higgsfield says GPT-6 Astra enables AI videos over 10 minutes. See current API rates, credit-cost limits, voice-to-video status and what to test.
Read →Anthropic Claude Abuse Rule: Pricing Impact (Oct 2026)
Anthropic's new Claude abuse rule takes effect November 12. See what it covers, what it excludes, and why API pricing is unchanged.
Read →Grok Bot Shared Bank Balances in Slack — Pricing Impact
A Grok Bot CFO agent sent a personal bank audit to company Slack. What happened, what remains unverified, and the real cost for buyers.
Read →All Major Providers
Provider pages give each pricing table more room than the homepage can.
Why Trust AI Pricing Guru?
Daily price checks
We check provider pricing pages every day. When a listed price moves, the tables should move with it.
Sources first
The tables show the rates first, so you don't have to start with a vendor blog post or sales page.
Built around API bills
Input price, output price, cache discounts, and usage caps all change the bill. The calculators are built around that math.
Frequently asked questions
Short answers about AI API pricing, subscriptions, and how we keep the data fresh.
What is the cheapest AI API in 2026?
The cheapest production AI API in today's 2026-10-19 snapshot is Llama 3.1 8B Instruct at $0.02 per million input tokens. Among flagship-class models, DeepSeek V4.1 Flash is currently lowest at $0.15 per million input tokens. We verify prices daily across 243 tracked models.
How does AI API pricing compare across OpenAI, Anthropic, Google, and DeepSeek?
The live table covers 243 models across 19 providers, including OpenAI, Anthropic, Google, DeepSeek, Mistral, xAI, Cohere, Groq, Together AI, Perplexity, and Meta Llama hosts. Compare them side-by-side on the API token pricing page with input, output, cached, and batch rates per million tokens.
How often are AI API prices updated on this site?
Pricing is verified and synced daily by an automated pipeline that checks provider-owned pricing sources against our maintained model records before publishing. New models and price changes typically appear here within hours of the provider's announcement. The current snapshot is from 2026-10-19.
Should I pay per token via API or subscribe to ChatGPT Plus or Claude Pro?
It depends on volume and usage limits. Current examples are ChatGPT Plus ($20.00/month), Claude Pro ($20.00/month), Google AI Pro ($19.99/month). API access is metered instead of bundled, so use the Subscription vs API calculator to find the break-even point for your workload across all 31 tracked plans.
How do I calculate the cost of AI tokens for my project?
Multiply your input tokens by the model's input price per 1M, then add your output tokens times the output price per 1M. For example, GPT-6.1 Sol at $2.00 input / $10.00 output per 1M tokens costs $0.002 per 1K input tokens plus $0.01 per 1K output tokens. The Token Cost Calculator does this automatically for any tracked model.
Can I download the AI API pricing data as JSON?
Yes. The full live dataset is published as a free, daily-updated JSON file at /api/pricing.json. Schema is { id, name, family, provider, pricing: { inputPerM, cachedInputPerM, outputPerM }, status } for all 243 tracked models. The AI Pricing Guru dataset is provided for informational use. You may use it for personal, editorial, research, educational, and internal business purposes with appropriate attribution to AI Pricing Guru. Commercial redistribution, resale, republishing at scale, inclusion in a competing public dataset/API, or use as the primary data source for a commercial pricing-comparison product requires prior written permission.
Ready to compare?
Find the best AI model for your budget in seconds.