Google
Gemini 2.5 Flash-Lite
High-volume extraction, routing, and simple multimodal tasks
- Input
- $0.1
- Output
- $0.4
- Context
- 1.05M
Recursive AI cost intelligence
Build AI workflows that improve with every run
Tokens.io turns live pricing, context windows, cache economics, and workload fit into a recursive self-improvement loop for model selection, prompt design, and routing decisions.
OpenAI Anthropic Gemini Mistral xAI DeepSeek Cohere
Loop inputLiveprovider-backed signals
Fit range64K-1.05Mcontext tracked by model
Routing field7major providers mapped
FreshnessJul 6, 2026automated source monitor
Recursive improvement loop
Every run should teach the next run where to spend less
Tokens.io is the measurement layer for AI operations: live prices, context fit, cache economics, and workload signals feed back into better prompts, model choices, and routing policy.
- Measure the cost and context footprint of each workload
- Route to the cheapest model that still fits the job
- Observe drift, cache misses, pricing changes, and spend spikes
- Improve the next policy, prompt, route, or model substitution
AI model directory
The live reference for model economics
Provider, model ID, pricing, context, source link, and workload fit in one browsable directory built for fast decisions.
OpenAI
GPT-5.4 nano
Classification, extraction, ranking, and sub-agent workloads
- Input
- $0.2
- Output
- $1.25
- Context
- 400K
DeepSeek
DeepSeek Chat
Low-cost chat, summarization, and extraction
- Input
- $0.27
- Output
- $1.1
- Context
- 64K
Mistral
Mistral Large 3
Open-weight multimodal workloads and self-hosting evaluation
- Input
- $0.5
- Output
- $1.5
- Context
- 256K
Google
Gemini 2.5 Flash
Large-scale, low-latency multimodal processing
- Input
- $0.3
- Output
- $2.5
- Context
- 1.05M
DeepSeek
DeepSeek Reasoner
Cost-sensitive reasoning workloads
- Input
- $0.55
- Output
- $2.19
- Context
- 64K
xAI
Grok Build 0.1
Agentic coding workflows on xAI
- Input
- $1
- Output
- $2
- Context
- 256K
xAI
Grok 4.3
General chat and reasoning outside code, audio, image, and video
- Input
- $1.25
- Output
- $2.5
- Context
- 1M
OpenAI
GPT-5.4 mini
Coding, computer-use, subagents, and efficient production agents
- Input
- $0.75
- Output
- $4.5
- Context
- 400K
Anthropic
Claude Haiku 4.5
Fast Claude workloads with near-frontier intelligence
- Input
- $1
- Output
- $5
- Context
- provider tier
Google
Gemini 2.5 Pro
Complex reasoning, coding, and long multimodal context
- Input
- $1.25
- Output
- $10
- Context
- 1.05M
Anthropic
Claude Sonnet 5
Balanced speed, intelligence, and production-agent cost
- Input
- $2
- Output
- $10
- Context
- 1M
Provider-sourced intelligence
Live model pricing table
Prices are USD per 1M tokens unless noted. Rows are normalized from provider sources and updated by Tokens.io monitors.
| Model | Provider | Input / 1M | Cached | Output / 1M | Context | Source |
|---|---|---|---|---|---|---|
| Gemini 2.5 Flash-Lite gemini-2.5-flash-lite | $0.1 | $0.01 | $0.4 | 1.05M | Gemini pricing | |
| GPT-5.4 nano gpt-5.4-nano | OpenAI | $0.2 | $0.02 | $1.25 | 400K | OpenAI pricing |
| DeepSeek Chat deepseek-chat | DeepSeek | $0.27 | $0.07 | $1.1 | 64K | DeepSeek pricing |
| Mistral Large 3 mistral-large-2512 | Mistral | $0.5 | - | $1.5 | 256K | Mistral model card |
| Gemini 2.5 Flash gemini-2.5-flash | $0.3 | $0.03 | $2.5 | 1.05M | Gemini pricing | |
| DeepSeek Reasoner deepseek-reasoner | DeepSeek | $0.55 | $0.14 | $2.19 | 64K | DeepSeek pricing |
| Grok Build 0.1 grok-build-0.1 | xAI | $1 | - | $2 | 256K | xAI models |
| Grok 4.3 grok-4.3 | xAI | $1.25 | - | $2.5 | 1M | xAI models |
| GPT-5.4 mini gpt-5.4-mini | OpenAI | $0.75 | $0.075 | $4.5 | 400K | OpenAI pricing |
| Claude Haiku 4.5 claude-haiku-4-5-20251001 | Anthropic | $1 | $0.1 | $5 | provider tier | Claude pricing |
| Gemini 2.5 Pro gemini-2.5-pro | $1.25 | $0.125 | $10 | 1.05M | Gemini pricing | |
| Claude Sonnet 5 claude-sonnet-5 | Anthropic | $2 | $0.2 | $10 | 1M | Claude pricing |
Instant calculator
Forecast monthly LLM spend before it becomes policy
Estimate input, output, and cache-hit economics before a workflow scales across users, agents, or teams.
- Separate input and output pricing
- Cache-aware estimates
- Sorted recommendations that can feed the next routing decision
Monthly input -
Monthly output -
Cheapest route -
| Model | Provider | Input | Output | Est. monthly |
|---|
Cost watchlist
Find the next substitution before the bill asks for it
Sorted by a simple Tokens.io blended-cost estimate: 70% input and 30% output. Use it to spot candidate routes, then validate quality, latency, and policy fit.
| # | Model | Provider | Blend / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | Gemini 2.5 Flash-Lite Volume | $0.19 | 1.05M | Gemini pricing | |
| 2 | GPT-5.4 nano Volume reasoning | OpenAI | $0.515 | 400K | OpenAI pricing |
| 3 | DeepSeek Chat Low-cost chat | DeepSeek | $0.519 | 64K | DeepSeek pricing |
| 4 | Mistral Large 3 Open-weight frontier | Mistral | $0.8 | 256K | Mistral model card |
| 5 | Gemini 2.5 Flash Fast multimodal | $0.96 | 1.05M | Gemini pricing | |
| 6 | DeepSeek Reasoner Reasoning value | DeepSeek | $1.042 | 64K | DeepSeek pricing |
| 7 | Grok Build 0.1 Coding | xAI | $1.3 | 256K | xAI models |
| 8 | Grok 4.3 General reasoning | xAI | $1.625 | 1M | xAI models |
Characters-
Words-
Estimated tokens-
Approx. input cost on GPT-5.4 nano-
Tokenizer tool 
Turn prompts, drafts, and agent traces into improvement signals
This lightweight tokenizer gives a planning estimate for text workloads so long prompts, repeated context, and agent traces can be improved before they scale.
Context window tracker
Know when context changes the next architecture decision
Long-context models change retrieval, agent memory, batch analysis, prompt packing, and the route that should run next.
| Model | Provider | Context window | Max output | Best for |
|---|---|---|---|---|
| GPT-5.4gpt-5.4 | OpenAI | 1.05M | 128K | Complex professional work at lower cost than GPT-5.5 |
| GPT-5.5gpt-5.5 | OpenAI | 1.05M | 128K | Hard reasoning, coding, and professional work |
| Gemini 2.5 Flash-Litegemini-2.5-flash-lite | 1.05M | 65.5K | High-volume extraction, routing, and simple multimodal tasks | |
| Gemini 2.5 Flashgemini-2.5-flash | 1.05M | 65.5K | Large-scale, low-latency multimodal processing | |
| Gemini 2.5 Progemini-2.5-pro | 1.05M | 65.5K | Complex reasoning, coding, and long multimodal context | |
| Grok 4.3grok-4.3 | xAI | 1M | - | General chat and reasoning outside code, audio, image, and video |
| Claude Sonnet 5claude-sonnet-5 | Anthropic | 1M | - | Balanced speed, intelligence, and production-agent cost |
| Claude Opus 4.8claude-opus-4-8 | Anthropic | 1M | - | Complex agentic coding and enterprise work |
| Claude Fable 5claude-fable-5 | Anthropic | 1M | - | Anthropic highest-capability work and long-running agents |
| GPT-5.4 nanogpt-5.4-nano | OpenAI | 400K | 128K | Classification, extraction, ranking, and sub-agent workloads |
Compare models
Pick the cheapest model that still fits the next job
Start with cost, context, and use-case fit, then let real workload results sharpen the next model substitution.
Google
Gemini 2.5 Flash-Lite
- Input
- $0.1
- Output
- $0.4
- Context
- 1.05M
High-volume extraction, routing, and simple multimodal tasks
OpenAI
GPT-5.4 nano
- Input
- $0.2
- Output
- $1.25
- Context
- 400K
Classification, extraction, ranking, and sub-agent workloads
DeepSeek
DeepSeek Chat
- Input
- $0.27
- Output
- $1.1
- Context
- 64K
Low-cost chat, summarization, and extraction
Mistral
Mistral Large 3
- Input
- $0.5
- Output
- $1.5
- Context
- 256K
Open-weight multimodal workloads and self-hosting evaluation
Google
Gemini 2.5 Flash
- Input
- $0.3
- Output
- $2.5
- Context
- 1.05M
Large-scale, low-latency multimodal processing
DeepSeek
DeepSeek Reasoner
- Input
- $0.55
- Output
- $2.19
- Context
- 64K
Cost-sensitive reasoning workloads
xAI
Grok Build 0.1
- Input
- $1
- Output
- $2
- Context
- 256K
Agentic coding workflows on xAI
xAI
Grok 4.3
- Input
- $1.25
- Output
- $2.5
- Context
- 1M
General chat and reasoning outside code, audio, image, and video
Provider pages
Free reference pages for every major AI provider
Provider cards summarize tracked models and link back to the underlying source docs for verification.
Anthropic
- Tracked models
- 4
- Lowest blended cost
- $2.2
- Largest context
- 1M
Cohere
- Tracked models
- 1
- Lowest blended cost
- $4.75
- Largest context
- 256K
DeepSeek
- Tracked models
- 2
- Lowest blended cost
- $0.519
- Largest context
- 64K
- Tracked models
- 3
- Lowest blended cost
- $0.19
- Largest context
- 1.05M
Mistral
- Tracked models
- 1
- Lowest blended cost
- $0.8
- Largest context
- 256K
OpenAI
- Tracked models
- 4
- Lowest blended cost
- $0.515
- Largest context
- 1.05M
xAI
- Tracked models
- 2
- Lowest blended cost
- $1.3
- Largest context
- 1M
Source intelligence
What changed in the public reference
The public layer is built for real-time lookup. Historical archives, alerts, feeds, and webhooks will make those changes actionable inside the machine-readable layer.
Provider-sourced pricing pass
Current rows now point to official provider docs for input, cached input, output, context, and model ID fields.
Unsourced demand claims removed
The public table now ranks by Tokens.io blended-cost estimate until usage or market-share methodology is ready.
Tiered pricing noted
Rows use standard public text-token rates where providers publish multiple batch, flex, priority, or long-context tiers.
Claude intro pricing flagged
Claude Sonnet 5 is shown at Anthropic introductory pricing, which Anthropic lists through August 31, 2026.
Live index clarified
Live public lookup stays free; API access, alerts, exports, history, and team workflows are positioned as the future machine layer.
Educational explainers
Make token economics legible to everyone using AI
Short, evergreen explainers build search authority and help buyers understand why token spend matters.
What is a token?
A token is the billing and processing unit language models use for text, code, and structured data.
Input vs output tokens
Input tokens are what you send to a model; output tokens are what the model generates back. They are often priced differently.
How cached input changes cost
Prompt caching can reduce repeat input cost when providers can reuse stable context across requests.
How to estimate AI costs
Multiply monthly input and output tokens by each model price, then include cache, tool, and priority surcharges.
Recursive TokenOps
From free lookup to a self-improving AI operating layer
The public site starts the loop with live intelligence. Teams pay when they need the same signals inside APIs, dashboards, alerts, exports, automation, and internal routing systems.
Measure every run
See cost, tokens, cache behavior, and context footprint across products, teams, and workflows.
Route with intent
Move work to the cheapest model that still fits quality, latency, context, and policy targets.
Observe drift
Catch price changes, prompt bloat, runaway agents, and cache misses as soon as they appear.
Improve policy
Feed each result back into better prompts, substitutions, routing rules, and budgets.
Developer data access
Machine-readable intelligence closes the loop
The current reference stays free. API plans will give apps, dashboards, alerts, exports, and routing layers the same live signals production systems need to improve decisions automatically.
Ask for early access when you want Tokens.io data inside your product, dashboard, agent stack, or routing workflow.