bladelabs/functiongemma-270m-router

FunctionGemma 270M Router

Edge-class banking tool router: 20 tools, 4 permission tiers. 270M parameters — small enough to run anywhere.

tool-calling

eval receipt

997-row held-out split

metric	base (untuned)	Blade tuned	120B teacher
tool selection	47.2%	98.2%	56.6%
argument validity	65.2%	86.6%	86.6%
permission recall	0%	100%	32.7%
permission violations	121	0	0

safety

Zero above-permission tool calls across the entire eval.

quickstart

OpenAI-compatible. Point your existing client at https://api.bladelabs.io/v1.

curl

curl https://api.bladelabs.io/v1/chat/completions \
  -H "Authorization: Bearer blade_sk_..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "bladelabs/functiongemma-270m-router",
    "messages": [{"role": "user", "content": "Transfer $50 to Alice"}],
    "tools": [ /* your tool schema */ ]
  }'

python

from openai import OpenAI

client = OpenAI(
    base_url="https://api.bladelabs.io/v1",
    api_key="blade_sk_...",
)

resp = client.chat.completions.create(
    model="bladelabs/functiongemma-270m-router",
    messages=[{"role": "user", "content": "Transfer $50 to Alice"}],
    tools=[ ],  # your tool schema
)

print(resp.choices[0].message.tool_calls)

Need a key? Create one in the dashboard →

pricing

$2 / 1M tokens

Blended prompt + completion. Metered per token, billed monthly.

context window 32k

versions

v1Early access