Best models

Best AI models for agents and tool use

Every model in the live reference catalogue with tool calls, structured output, ranked by the agentic index (Artificial Analysis, 0-100) with vendor list prices per million tokens. 12 qualify today.

Strongest

Claude Fable 5.1

agentic index 58 · $20.0 blended

Best value

GLM 5.3 Flash

51 points for $0.24

Cheapest

GLM 5.3 Flash

$0.24 blended · index 51

Ranked by the agentic index

Strongest first; price breaks ties. Value is index points per blended dollar. Click a model for its full specification, or compare two side by side.

#	Model	Agentic index	Input / 1M	Output / 1M	Context	Value	Compare
1	Claude Fable 5.1Anthropic	58	$10.0	$50.0	1M	3
2	Claude Opus 5Anthropic	57	$5.00	$25.0	1M	6	vs Claude Fable 5.1
3	Qwen3.8 Max (0902)Qwen	56	$2.00	$6.00	1M	19	vs Claude Fable 5.1
4	Muse Spark 1.3Meta	56	$1.25	$4.25	1.0M	28	vs Claude Fable 5.1
5	GLM 5.3Z.ai	53	$0.08	$1.40	1.0M	130	vs Claude Fable 5.1
6	Grok 4.6xAI	53	$2.00	$6.00	500K	18	vs Claude Fable 5.1
7	GPT-6 AstraOpenAI	51	$10.0	$50.0	1.1M	3	vs Claude Fable 5.1
8	GLM 5.3 FlashZ.ai	51	$0.15	$0.50	1.0M	214	vs Claude Fable 5.1
9	Claude Fable 5Anthropic	51	$10.0	$50.0	1M	3	vs Claude Fable 5.1
10	GPT-5.6 SolOpenAI	50	$4.00	$20.0	1.1M	6	vs Claude Fable 5.1
11	Qwen3.8 2.4T A95BQwen	50	$2.00	$6.00	1.0M	17	vs Claude Fable 5.1
12	Kimi K3Moonshot AI	50	$0.24	$11.9	1.0M	16	vs Claude Fable 5.1

What to weigh for agents and tool use

An agent's model is asked to do three things a chat model is not: decide which tool to call, format the call exactly, and keep going for ten or twenty steps without losing the thread. The agentic index measures tool use and multi-step task completion; tool calls and structured output are required because an agent that emits a malformed call stalls.

Agents also multiply tokens - every step re-reads the conversation - so the price column compounds. A model that costs twice as much per token costs twice as much per step and the steps add up; the value ranking below (index per dollar) is often the better guide for the steps that are not the hard one.

The router treats each step as its own request: the planning step and the final answer go to the strongest qualifying model, the routine tool-formatting steps to the cheapest one that scores well on the agentic index. Resource limits cap the steps and the spend so a looping agent stops instead of running up a bill, and the trace of a run shows which model took which step and why.

The rule, in OpenSmartRoute

Declare the models as targets and let the rule send the hard requests to the strongest one and the rest to the best value; the trace names the rule that fired.

yaml

# rules.yaml - agents: the strongest qualifying model for the hard requests, the best value for the rest
rules:
  - name: agents-hard
    when: { actions: [action, planning], needs_tools: true, min_complexity: 0.4 }
    prefer: [claude-fable-5.1]
    weight: 0.8
  - name: agents-default
    when: { max_complexity: 0.45 }
    prefer: [glm-5.3-flash]
    weight: 0.6

Try a request Price a workload

Frequently asked

Which model is best for building AI agents?: The top of the agentic ranking above - computed from the live catalogue - among models that support tool calls and structured output.
Can an agent use several models?: Yes, and it should. Each step is a routing decision: the hard planning step gets the strongest model, the tool-formatting steps a cheap one. The router does this with one rule.
How do I stop an agent from overspending?: Resource limits on the agent target cap steps, time and cost per request; budgets cap the target per day. Both are declared in the catalogue, not in your code.

Best models for other tasks

For the business

What it means for which models you pay for.

Catalogue of every vendor
The right model, not the famous one
Price, context, abilities and benchmark indexes side by side, with the model that fits each task named.
The rankings
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
Per million tokens, every model
Prices you can compare
The catalogue prints input and output prices and what three typical monthly workloads cost on each.
The model catalogue

Which LLM is the cheapest for our workload?: The best-for pages rank the live catalogue per task by the index that measures it, and the router picks the cheapest model that meets your quality rule for each request.
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 often do model prices change?: The catalogue is refreshed from the vendors' price lists; each model page shows the date of its data.

Let the router pick the agents and tool use model per request.

The ranking above is where to start; the router finishes the job - the strongest qualifying model for the hard requests, the best value for the rest, and every decision explained.

Create a free account Price a workload

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

Model

Agentic index

Input / 1M

Output / 1M

Context

Value

Compare

Claude Fable 5.1Anthropic

$10.0

$50.0

Claude Opus 5Anthropic

$5.00

$25.0

vs Claude Fable 5.1

Qwen3.8 Max (0902)Qwen

$2.00

$6.00

vs Claude Fable 5.1

Muse Spark 1.3Meta

$1.25

$4.25

1.0M

vs Claude Fable 5.1

GLM 5.3Z.ai

$0.08

$1.40

1.0M

130

vs Claude Fable 5.1

Grok 4.6xAI

$2.00

$6.00

500K

vs Claude Fable 5.1

GPT-6 AstraOpenAI

$10.0

$50.0

1.1M

vs Claude Fable 5.1

GLM 5.3 FlashZ.ai

$0.15

$0.50

1.0M

214

vs Claude Fable 5.1

Claude Fable 5Anthropic

$10.0

$50.0

vs Claude Fable 5.1

GPT-5.6 SolOpenAI

$4.00

$20.0

1.1M

vs Claude Fable 5.1

Qwen3.8 2.4T A95BQwen

$2.00

$6.00

1.0M

vs Claude Fable 5.1

Kimi K3Moonshot AI

$0.24

$11.9

1.0M

vs Claude Fable 5.1

What to weigh for agents and tool use

The rule, in OpenSmartRoute

Declare the models as targets and let the rule send the hard requests to the strongest one and the rest to the best value; the trace names the rule that fired.

yaml

# rules.yaml - agents: the strongest qualifying model for the hard requests, the best value for the rest
rules:
  - name: agents-hard
    when: { actions: [action, planning], needs_tools: true, min_complexity: 0.4 }
    prefer: [claude-fable-5.1]
    weight: 0.8
  - name: agents-default
    when: { max_complexity: 0.45 }
    prefer: [glm-5.3-flash]
    weight: 0.6

Frequently asked

Which model is best for building AI agents?

The top of the agentic ranking above - computed from the live catalogue - among models that support tool calls and structured output.

Can an agent use several models?

Yes, and it should. Each step is a routing decision: the hard planning step gets the strongest model, the tool-formatting steps a cheap one. The router does this with one rule.

How do I stop an agent from overspending?

Resource limits on the agent target cap steps, time and cost per request; budgets cap the target per day. Both are declared in the catalogue, not in your code.

Best AI models for agents and tool use

Ranked by the agentic index

What to weigh for agents and tool use

The rule, in OpenSmartRoute

Frequently asked

Related

Best models for other tasks

What it means for which models you pay for.

The right model, not the famous one

Lower cost per request

Prices you can compare

Let the router pick the agents and tool use model per request.

Best AI models for agents and tool use

Ranked by the agentic index

What to weigh for agents and tool use

The rule, in OpenSmartRoute

Frequently asked

Related

Best models for other tasks

What it means for which models you pay for.

The right model, not the famous one

Lower cost per request

Prices you can compare

Let the router pick the agents and tool use model per request.