smolagents 1.26.0: our test log

smolagents 1.26.0 passed our one-tool task on a local 3B model: 6 of 6 trials; cold uv install 1.5 s.

Result at a glance

What we found

• 6 of 6 task trials passed: the model's tool call reached the framework, the function ran with 17 and 25, and the final answer contained 42.
• Cold install took 1.5 s with uv and 5.6 s with pip, and left 54 MB in 28 packages (uv).
• We ran both agent styles: ToolCallingAgent (the model emits JSON tool calls) and CodeAgent (the model writes Python that calls the tool). Per-style pass counts are in the step table.
• CodeAgent executes model-written Python in a local interpreter; we did not test its sandbox options.

What we ran, step by step

# Install with uv (cold, no cache)
uv venv -q -p 3.13 work/py/smolagents && VIRTUAL_ENV=work/py/smolagents uv pip install --no-cache -q smolagents

# Install with pip (cold, no cache)
uv venv -q --seed -p 3.13 work/pip/smolagents && work/pip/smolagents/bin/pip install --no-cache-dir -q smolagents

# Task command (all trials)
OTEL_SDK_DISABLED=true work/py/smolagents/bin/python tasks/smolagents_agent.py

Raw output (excerpts)

from common import *
from smolagents import ToolCallingAgent, OpenAIServerModel, tool
@tool
def add_numbers_sm(a: int, b: int) -> int:
    """Add two integers and return the sum.

    Args:
        a: first integer
        b: second integer
    """
    return add_numbers(a, b)
t = add_numbers_sm
agent = ToolCallingAgent(tools=[t], model=OpenAIServerModel(model_id=MODEL, api_base=BASE, api_key=KEY), verbosity_level=0, max_steps=4)
report(agent.run(TASK))

import json, os
BASE = "http://127.0.0.1:11434/v1"
MODEL = os.environ.get("MODEL", "qwen2.5-3b")
KEY = "ollama"
TASK = "Use the add_numbers tool to compute 17 + 25, then tell me the result."
CALLS = []
def add_numbers(a: int, b: int) -> int:
    """Add two integers and return the sum."""
    CALLS.append((a, b))
    return int(a) + int(b)
def report(final):
    ok = any({int(x), int(y)} == {17, 25} for x, y in CALLS) and "42" in str(final)
    print("RESULT_JSON " + json.dumps({"calls": CALLS, "final": str(final)[:300], "ok": ok}))

RESULT_JSON {"calls": [[17, 25]], "final": "The result of adding 17 and 25 is 42.", "ok": true}

from common import *
from common import add_numbers as _add
from smolagents import CodeAgent, OpenAIServerModel, tool
@tool
def add_numbers(a: int, b: int) -> int:
    """Add two integers and return the sum.

    Args:
        a: first integer
        b: second integer
    """
    return _add(a, b)
t = add_numbers
agent = CodeAgent(tools=[t], model=OpenAIServerModel(model_id=MODEL, api_base=BASE, api_key=KEY), verbosity_level=0, max_steps=4)
report(agent.run(TASK))

Gotchas and things that broke

• The tool function needs a docstring that describes every argument, or smolagents raises DocstringParsingException (Cannot generate JSON schema ... the docstring has no description for the argument) at definition time.
• On a bare pip install smolagents, constructing OpenAIServerModel raises ModuleNotFoundError: Please install 'openai' extra (smolagents[openai]). The install size and time above are for the bare package; we did not time the extra separately.

What we did not test

• Any vendor-hosted model: everything ran on a local 3B model, so tool-call reliability with frontier models is not measured here.
• Multi-agent flows, memory, long conversations, streaming, tracing and cloud deployment.
• Anything beyond one tool and one question; three trials is a smoke test, not a benchmark.
• The CodeAgent sandboxes (Docker, E2B, Modal) and the hub tools.

Reproduce it

Versions change. These commands install whatever is latest on the day you run them unless a version is pinned, so expect different results later. The version we measured is in the table above.

# 1. Serve any local model that supports tool calling behind an OpenAI-compatible endpoint, registered as qwen2.5-3b:
#    http://127.0.0.1:11434/v1  (we used Ollama with a Qwen2.5-3B-Instruct Q4_K_M GGUF)
uv venv -p 3.13 .venv && VIRTUAL_ENV=.venv uv pip install smolagents
# 2. Save common.py (above) and the task script next to each other, then:
.venv/bin/python smolagents_agent.py

How to read this test

Every number on this page was measured on this one machine, which also ran other jobs, so timings are indicative and vary run to run. A pass means our small task worked once or as stated, a fail means it did not work as we ran it; neither says how the tool behaves with a vendor's frontier model, on a GPU, or at scale. Where a tool could not run without an account or a GPU, we say so instead of guessing.

See the testing rig for the hardware, the local model and the rules we follow, and the test hub for every other test.

Related tests: agent frameworks and sdks

• Vercel AI SDK: Vercel AI SDK passed our one-tool task 3 of 3 on Node 20 (ai 6.0.315) and 3 of 3 on Node 24 (ai 7.0.153); the install is 26 to 27 MB.
• Agno: Agno 3.1.10 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 1.8 s.
• AutoGen: AutoGen 0.7.5 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 1.9 s.
• CrewAI: CrewAI 1.15.30 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 8.4 s.
• DSPy: DSPy 3.4.1 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 3.7 s.
• Google ADK (Python): Google ADK (Python) 2.13.0 passed our one-tool task on a local 3B model: 3 of 3 trials; cold uv install 1.7 s.

All test logs · How we test

FAQ

Sources

• smolagents repository (fetched 2026-10-25)

Related

• All test logs
• Methodology and testing rig
• Study: install size and first-task results across 16 frameworks
• smolagents facts and sources