How much can we save on our LLM bill?

On the public benchmark the router saved 89% against sending everything to the reference model. Import last month's provider bill on the bill page and the estimate is computed on your own traffic before you change a line of code.

How quickly does routing pay back?

There is nothing to pay back: the free plan and the open-source router cost nothing to start, and the first routed request already carries its own cost against the baseline.

How do we keep AI spend under control by team?

Each workspace, each tenant and each API key can carry a daily or monthly budget and a request quota; the router enforces them and the dashboard shows spend against the line as it happens.

Cost estimator

Know what a request costs before you send it.

Paste a prompt. The router estimates tokens, scores every target and quotes each one with its price split into input and output, expected quality and latency, then names the recommended, cheapest capable, best-quality and fastest picks.

Your request

Nothing is executed and the prompt is not stored. Tokens are estimated from the text; costs use each model's list price.

Know the price before you send

The router reads your prompt, estimates tokens, scores every model, agent and skill in the catalogue and quotes each one with its expected quality and latency.

Recommended models

What the router picks per use case

Live quotes of representative prompts against this deployment's catalogue, recomputed every minute. Rankings follow the targets, list prices and learned quality, not a hand-maintained list.

Everyday assistant

16 candidates

Mid-tier model

GPT-4.1 Minillm

per request: $0.00065
routing quality: Routing quality score: the router's own estimate of how well this target would answer this prompt (0-1), from its quality prior and learned outcomes - not an independent benchmark.: 0.69
est. latency: Estimated latency: the target's declared or observed time to answer, as the router weighs it for this prompt - not a standardised benchmark.: 900.0 ms

Why: domain fit 50% · complexity fit 100% · quality prior 0.75

Cheapestthis pick
Best qualitythis pick
Fastestthis pick

Coding and refactoring

16 candidates

Coding agent

coding-agentagent

per request: $0.0029
routing quality: Routing quality score: the router's own estimate of how well this target would answer this prompt (0-1), from its quality prior and learned outcomes - not an independent benchmark.: 0.76
est. latency: Estimated latency: the target's declared or observed time to answer, as the router weighs it for this prompt - not a standardised benchmark.: 6.00 s

Why: rule 'hard-coding-to-agent' prefers · domain fit 100% · action fit 50%

Analysis and reasoning

16 candidates

Mid-tier model

GPT-4.1 Minillm

per request: $0.0005
routing quality: Routing quality score: the router's own estimate of how well this target would answer this prompt (0-1), from its quality prior and learned outcomes - not an independent benchmark.: 0.69
est. latency: Estimated latency: the target's declared or observed time to answer, as the router weighs it for this prompt - not a standardised benchmark.: 900.0 ms

Why: domain fit 50% · complexity fit 100% · quality prior 0.75

Cheapestgpt-oss-20b (self-hosted)$0.40/1M
Best qualitythis pick
Fastestthis pick

Summaries and rewriting

16 candidates

Summarization skill

skill-summarizeskill

per request: $0.00017
routing quality: Routing quality score: the router's own estimate of how well this target would answer this prompt (0-1), from its quality prior and learned outcomes - not an independent benchmark.: 0.74
est. latency: Estimated latency: the target's declared or observed time to answer, as the router weighs it for this prompt - not a standardised benchmark.: 1.20 s

Why: rule 'legal-persona' prefers · domain fit 100% · complexity fit 100%

CheapestOn-prem private model$1.00/1M
Best qualityUthis pick
FastestOn-prem private model$1.00/1M

Extraction and classification

16 candidates

Document and image reader (OCR)

qwen3-vl-4b-instructllm

per request: $0.000017
routing quality: Routing quality score: the router's own estimate of how well this target would answer this prompt (0-1), from its quality prior and learned outcomes - not an independent benchmark.: 0.71
est. latency: Estimated latency: the target's declared or observed time to answer, as the router weighs it for this prompt - not a standardised benchmark.: 1.50 s

Why: domain fit 33% · complexity fit 100% · quality prior 0.74

CheapestUthis pick
Best qualityUthis pick
FastestSmall fast model$0.20/1M

Sensitive data

12 candidates

On-prem private model

Llama 3.3 70B Instructllm

per request: $0.000067
routing quality: Routing quality score: the router's own estimate of how well this target would answer this prompt (0-1), from its quality prior and learned outcomes - not an independent benchmark.: 0.88
est. latency: Estimated latency: the target's declared or observed time to answer, as the router weighs it for this prompt - not a standardised benchmark.: 1.20 s

Why: rule 'pii-stays-onprem' prefers · domain fit 100% · complexity fit 100%

Cheapestthis pick
Best qualitythis pick
Fastestthis pick

Most recommended across use cases

1
Mid-tier model
router's pick for 2 use cases · cheapest 1x · best quality 2x
llm
2
On-prem private model
router's pick for 1 use case · cheapest 2x · best quality 1x
llm
3U
Document and image reader (OCR)
router's pick for 1 use case · cheapest 1x · best quality 1x
llm
4U
Coding agent
router's pick for 1 use case · best quality 1x
agent
5U
Summarization skill
router's pick for 1 use case · best quality 1x
skill
6
Frontier reasoning model
2 placements · cheapest 1x
llm

Before the bill, not after

A quote is a number you can put in front of someone before the money is spent.

Every figure on this page is the same one the router uses when it decides: list prices per token, the tokens your prompt will take, the quality each target has earned on requests like it. Paste the prompt your team sends most, and the difference between the model you use today and the cheapest one that would do is on the screen in a second.

Photo: Jakub Zerdzicki

Quote from code

The same quote is one call away: anonymous quotes are rate limited per client, keyed quotes apply your tenant rules and show up in your usage.

POST /api/v1/estimatebash

curl -s /api/v1/estimate -H "Authorization: Bearer $OSR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Summarise this contract...", "monthly_requests": 50000}'

Or from the SDK and your IDE

osr estimate quotes against a local catalogue and the MCP server exposes estimate and recommend tools to VS Code, Cursor and Claude.

opensmartroute.estimatepython

from opensmartroute.estimate import estimate

q = estimate(router, "Summarise this contract...", output_tokens=300)
print(q.recommended.name, q.recommended.cost_usd, q.savings_usd)

Set up the IDE integration under Integrations or read the MCP guide.

For the business

What it does to the AI bill.

89% less
Lower cost per request
On the 150-prompt benchmark the router spent 89% less than sending every prompt to the reference model, and the judged quality went up, not down.
Read the benchmark
From last month's bill
Your own estimate
The calculator works from your real volumes and models, not an assumed mix.
Calculate your savings
Budgets per workspace, tenant and key
A bill you can forecast
Set the daily or monthly line per workspace, per tenant and per API key; the router answers 429 at it instead of surprising finance at month end.
How budgets work

How much can we save on our LLM bill?: On the public benchmark the router saved 89% against sending everything to the reference model. Import last month's provider bill on the bill page and the estimate is computed on your own traffic before you change a line of code.
How quickly does routing pay back?: There is nothing to pay back: the free plan and the open-source router cost nothing to start, and the first routed request already carries its own cost against the baseline.
How do we keep AI spend under control by team?: Each workspace, each tenant and each API key can carry a daily or monthly budget and a request quota; the router enforces them and the dashboard shows spend against the line as it happens.

Quote first. Spend second.

A quote needs no account. When the number is right, the same request goes through the router with one more field - and comes back with the model, the price and the reason.

Quote a request now Create a free account

Free plan, no card. Fifteen thousand decisions a month with the full trace.

A quote is a number you can put in front of someone before the money is spent.

Quote from code

The same quote is one call away: anonymous quotes are rate limited per client, keyed quotes apply your tenant rules and show up in your usage.

POST /api/v1/estimatebash

curl -s /api/v1/estimate -H "Authorization: Bearer $OSR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Summarise this contract...", "monthly_requests": 50000}'

Or from the SDK and your IDE

osr estimate quotes against a local catalogue and the MCP server exposes estimate and recommend tools to VS Code, Cursor and Claude.

opensmartroute.estimatepython

from opensmartroute.estimate import estimate

q = estimate(router, "Summarise this contract...", output_tokens=300)
print(q.recommended.name, q.recommended.cost_usd, q.savings_usd)

Set up the IDE integration under Integrations or read the MCP guide.

Know what a request costs before you send it.

Your request

Know the price before you send

What the router picks per use case

Most recommended across use cases

A quote is a number you can put in front of someone before the money is spent.

Quote from code

Or from the SDK and your IDE

What it does to the AI bill.

Lower cost per request

Your own estimate

A bill you can forecast

Quote first. Spend second.

Know what a request costs before you send it.

Your request

Know the price before you send

What the router picks per use case

Most recommended across use cases

A quote is a number you can put in front of someone before the money is spent.

Quote from code

Or from the SDK and your IDE

What it does to the AI bill.

Lower cost per request

Your own estimate

A bill you can forecast

Quote first. Spend second.