Most AI tasks no longer need a frontier model Read launch post

More accurate tool calls than GPT-⁠6.1 Sol, 90% cheaper

A coding agent works through tool calls, from opening a file to running the tests. SIE serves GLM-5.3-Flash, an open model, behind an OpenAI-compatible API, so switching an agent means changing its base URL and model name.

SIE GLM-5.3-Flash

Tests pass
Task

Make the tests pass. Run them with uv run pytest -q. Do not change the tests.

  1. read invoice_totals/amounts.py

  2. read tests/test_amounts.py

  3. edit invoice_totals/amounts.py

    if cleaned.startswith("(") and cleaned.endswith(")"):    cleaned = "-" + cleaned[1:-1].strip()return Decimal(cleaned)
  4. bash uv run pytest -q

    3 passed in 0.00s

GPT-6 Luna invents a date; Claude Sonnet 5.5 makes a stray Write call

GPT-6 Luna (high reasoning)

Wrong value
Request, sent 21 October 2026
Get information about the pop concerts in New York for next month.
Declared function
  • concert_info_get

    date string, required

    Time frame to get the concert for.

Tool call returned
concert_info_get({  "date": "April 2026",  "genre": "Pop",  "location": "New York"})
Expected call
concert_info_get({  "date": "next month",  "genre": "Pop",  "location": "New York"})
Read the complete record

90% cheaper than GPT-⁠6.1 Sol, more accurate than GPT-⁠6 Luna

Tool calls right
600 calls ↑
Claude Sonnet 5.5
Claude Sonnet 5.5 92.0%
SIE GLM-5.3-Flash
SIE GLM-5.3-Flash 91.6%
Claude Haiku 5.5
Claude Haiku 5.5 91.2%
GPT-6 Luna
GPT-6 Luna 86.8%
GPT-6.1 Sol (high reasoning)
GPT-6.1 Sol (high reasoning) 84.3%
GPT-6.1 Sol (low reasoning)
GPT-6.1 Sol (low reasoning) 83.7%
Cost per 1,000 tool calls
USD ↓
GPT-6 Luna
$0.0321 GPT-6 Luna
SIE GLM-5.3-Flash
$0.0634 SIE GLM-5.3-Flash
Claude Haiku 5.5
$0.115 Claude Haiku 5.5
GPT-6.1 Sol (low reasoning)
$0.666 GPT-6.1 Sol (low reasoning)
GPT-6.1 Sol (high reasoning)
$0.703 GPT-6.1 Sol (high reasoning)
Claude Sonnet 5.5
Claude Sonnet 5.5 $2.27
Evaluation
Tested
600 requests from the Berkeley Function Calling Leaderboard, each with its functions
Scored
A call is right when its function and every argument match the benchmark's answer
Prices
OpenAI and Anthropic list prices; SIE at its Serverless price

Send the failing file and your tools, get back the edit that fixes it

import json
from sie_sdk import SIEClient
​
client = SIEClient(api_key="API keysk-sie-…")
​
tools = json.loads("""
[
{
"type": "function",
"function": {
"name": "read_file",
"description": "Read a file in the repository and return its contents.",
"parameters": {"type": "object", "properties": {"filePath": {"type": "string", "description": "Path of the file, relative to the repository root"}}, "required": ["filePath"], "additionalProperties": false}
}
},
{
"type": "function",
"function": {
"name": "edit",
"description": "Replace an exact string in a file with new text.",
"parameters": {"type": "object", "properties": {"filePath": {"type": "string", "description": "Path of the file, relative to the repository root"}, "oldString": {"type": "string", "description": "The exact text to replace"}, "newString": {"type": "string", "description": "The text to put in its place"}, "replaceAll": {"type": "boolean", "description": "Replace every occurrence (default false)"}}, "required": ["filePath", "oldString", "newString"], "additionalProperties": false}
}
}
]
""")
messages = [
{"role": "system", "content": "instructionYou are a coding agent working in a Python repository. Change files only with the tools you are given."},
{"role": "user", "content": "textinvoice_totals/amounts.py: ```python """Parse the amount column of an exported invoice into Decimal values.""" from decimal import Decimal def parse_amount(text: str) -> Decimal: """Turn one amount cell into a Decimal. Accepts "$1,234.50", "1234.5" and " 12.00 ". A credit is written with parentheses, "(12.00)", and is a negative amount. """ cleaned = text.strip().replace("$", "").replace(",", "") return Decimal(cleaned) def total(cells: list[str]) -> Decimal: """Net total of the amount column, credits subtracted.""" return sum((parse_amount(cell) for cell in cells), Decimal(0)) ``` tests/test_amounts.py fails: test_credit_in_parentheses_is_negative expects parse_amount("(12.00)") to equal Decimal("-12.00"). Fix parse_amount so a credit written in parentheses parses as a negative amount. Do not change the tests."},
]
result = client.chat_completions(
"modelzai-org/GLM-5.3-Flash:relay",
messages,
max_completion_tokens=max tokens4096,
extra_body={"temperature":0,"top_p":1},
tools=tools,
)
message = result["choices"][0]["message"]
if message.get("content"):
print(message["content"])
for call in message.get("tool_calls") or []:
args = call["function"]["arguments"]
try:
args = json.loads(args or "{}")
except json.JSONDecodeError:
pass
print(call["function"]["name"], args)
import { SIEClient, type ToolSpec } from '@superlinked/sie-sdk';

const client = new SIEClient(undefined, { apiKey: 'sk-sie-…' });

const tools: ToolSpec[] = [
  {
    "type": "function",
    "function": {
      "name": "read_file",
      "description": "Read a file in the repository and return its contents.",
      "parameters": {"type": "object", "properties": {"filePath": {"type": "string", "description": "Path of the file, relative to the repository root"}}, "required": ["filePath"], "additionalProperties": false}
    }
  },
  {
    "type": "function",
    "function": {
      "name": "edit",
      "description": "Replace an exact string in a file with new text.",
      "parameters": {"type": "object", "properties": {"filePath": {"type": "string", "description": "Path of the file, relative to the repository root"}, "oldString": {"type": "string", "description": "The exact text to replace"}, "newString": {"type": "string", "description": "The text to put in its place"}, "replaceAll": {"type": "boolean", "description": "Replace every occurrence (default false)"}}, "required": ["filePath", "oldString", "newString"], "additionalProperties": false}
    }
  }
];
const result = await client.chatCompletions({
  model: 'zai-org/GLM-5.3-Flash:relay',
  messages: [
    { role: 'system', content: "You are a coding agent working in a Python repository. Change files only with the tools you are given." },
    { role: 'user', content: "invoice_totals/amounts.py:\n\n```python\n\"\"\"Parse the amount column of an exported invoice into Decimal values.\"\"\"\n\nfrom decimal import Decimal\n\n\ndef parse_amount(text: str) -> Decimal:\n    \"\"\"Turn one amount cell into a Decimal.\n\n    Accepts \"$1,234.50\", \"1234.5\" and \"  12.00 \". A credit is written with\n    parentheses, \"(12.00)\", and is a negative amount.\n    \"\"\"\n    cleaned = text.strip().replace(\"$\", \"\").replace(\",\", \"\")\n    return Decimal(cleaned)\n\n\ndef total(cells: list[str]) -> Decimal:\n    \"\"\"Net total of the amount column, credits subtracted.\"\"\"\n    return sum((parse_amount(cell) for cell in cells), Decimal(0))\n```\n\ntests/test_amounts.py fails: test_credit_in_parentheses_is_negative expects parse_amount(\"(12.00)\") to equal Decimal(\"-12.00\"). Fix parse_amount so a credit written in parentheses parses as a negative amount. Do not change the tests." },
  ],
  max_completion_tokens: 4096,
  temperature: 0,
  top_p: 1,
  tools,
});
const message = result.choices[0]?.message;
if (message?.content) console.log(message.content);
for (const call of message?.tool_calls ?? []) {
  let args: unknown = call.function.arguments;
  try {
    args = JSON.parse(call.function.arguments || '{}');
  } catch {}
  console.log(call.function.name, args);
}
curl https://api.superlinked.com/v1/chat/completions \
  -H "Authorization: Bearer sk-sie-…" \
  -H "Content-Type: application/json" \
  -d "{\"model\":\"zai-org/GLM-5.3-Flash:relay\",\"messages\":[{\"role\":\"system\",\"content\":\"You are a coding agent working in a Python repository. Change files only with the tools you are given.\"},{\"role\":\"user\",\"content\":\"invoice_totals/amounts.py:\\n\\n\`\`\`python\\n\\\"\\\"\\\"Parse the amount column of an exported invoice into Decimal values.\\\"\\\"\\\"\\n\\nfrom decimal import Decimal\\n\\n\\ndef parse_amount(text: str) -> Decimal:\\n    \\\"\\\"\\\"Turn one amount cell into a Decimal.\\n\\n    Accepts \\\"\$1,234.50\\\", \\\"1234.5\\\" and \\\"  12.00 \\\". A credit is written with\\n    parentheses, \\\"(12.00)\\\", and is a negative amount.\\n    \\\"\\\"\\\"\\n    cleaned = text.strip().replace(\\\"\$\\\", \\\"\\\").replace(\\\",\\\", \\\"\\\")\\n    return Decimal(cleaned)\\n\\n\\ndef total(cells: list[str]) -> Decimal:\\n    \\\"\\\"\\\"Net total of the amount column, credits subtracted.\\\"\\\"\\\"\\n    return sum((parse_amount(cell) for cell in cells), Decimal(0))\\n\`\`\`\\n\\ntests/test_amounts.py fails: test_credit_in_parentheses_is_negative expects parse_amount(\\\"(12.00)\\\") to equal Decimal(\\\"-12.00\\\"). Fix parse_amount so a credit written in parentheses parses as a negative amount. Do not change the tests.\"}],\"max_completion_tokens\":4096,\"temperature\":0,\"top_p\":1,\"tools\":[{\"type\":\"function\",\"function\":{\"name\":\"read_file\",\"description\":\"Read a file in the repository and return its contents.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"filePath\":{\"type\":\"string\",\"description\":\"Path of the file, relative to the repository root\"}},\"required\":[\"filePath\"],\"additionalProperties\":false}}},{\"type\":\"function\",\"function\":{\"name\":\"edit\",\"description\":\"Replace an exact string in a file with new text.\",\"parameters\":{\"type\":\"object\",\"properties\":{\"filePath\":{\"type\":\"string\",\"description\":\"Path of the file, relative to the repository root\"},\"oldString\":{\"type\":\"string\",\"description\":\"The exact text to replace\"},\"newString\":{\"type\":\"string\",\"description\":\"The text to put in its place\"},\"replaceAll\":{\"type\":\"boolean\",\"description\":\"Replace every occurrence (default false)\"}},\"required\":[\"filePath\",\"oldString\",\"newString\"],\"additionalProperties\":false}}}]}"
Build the "Coding" capability into my app using the Superlinked Inference Engine (SIE).

Context
- SIE is an OpenAI-style inference API. Python SDK: `from sie_sdk import SIEClient`; TypeScript: `@superlinked/sie-sdk`.
- Base URL: https://api.superlinked.com (or my regional endpoint). Auth: Bearer key from env `SIE_API_KEY` (never hard-code it).
- Model: zai-org/GLM-5.3-Flash:relay (OpenAI-compatible endpoint: /v1/chat/completions). Keep the model id configurable.

Task
- Input: a source file and a request to change it, plus the coding agent's tools.
- Behaviour: return the tool calls the model makes to change the file: each call's function name and its parsed arguments
- Send one POST /v1/chat/completions request (SDK: chat_completions / chatCompletions) with an optional system message for instructions, the user's text as the user message, and max_completion_tokens 4096, temperature 0, top_p 1. Send the tools below unchanged as tools; send no tool_choice. Read any reply text from choices[0].message.content and each tool call from choices[0].message.tool_calls: its function.name, and its function.arguments, a JSON string to parse. Tools, sent as given: [{"type":"function","function":{"name":"read_file","description":"Read a file in the repository and return its contents.","parameters":{"type":"object","properties":{"filePath":{"type":"string","description":"Path of the file, relative to the repository root"}},"required":["filePath"],"additionalProperties":false}}},{"type":"function","function":{"name":"edit","description":"Replace an exact string in a file with new text.","parameters":{"type":"object","properties":{"filePath":{"type":"string","description":"Path of the file, relative to the repository root"},"oldString":{"type":"string","description":"The exact text to replace"},"newString":{"type":"string","description":"The text to put in its place"},"replaceAll":{"type":"boolean","description":"Replace every occurrence (default false)"}},"required":["filePath","oldString","newString"],"additionalProperties":false}}}]

Deliverables
- A typed client wrapper, an application-level function for this task, error handling for timeouts/empty input, and unit tests with a stubbed client.
- Wire it into my existing stack (ask me which framework if unclear) and add a short usage example.
Output
tool_calls[0].function.name
edit
arguments.filePath
invoice_totals/amounts.py
arguments.oldString
    cleaned = text.strip().replace("$", "").replace(",", "")
    return Decimal(cleaned)
arguments.newString
    cleaned = text.strip().replace("$", "").replace(",", "")
    if cleaned.startswith("(") and cleaned.endswith(")"):
        cleaned = "-" + cleaned[1:-1]
    return Decimal(cleaned)

Run your coding agent on Serverless or on your own cluster, down to one GPU

SIE Cluster

Kubernetes in your own cloud, backed by our engineers.

  • Terraform modules for your cloud
  • Apache-2.0, same engine as Serverless
  • GPU workers scale to zero between jobs
  • Separate worker pools for each tenant
Agent prompt Deploy SIE to our AWS account with the superlinked/sie/aws Terraform module. Docs: superlinked.com/docs/deployment.Deploy SIE to our Google Cloud project with the superlinked/sie/google Terraform module. Docs: superlinked.com/docs/deployment.Deploy SIE to our Azure subscription with the superlinked/sie/azure Terraform module. Docs: superlinked.com/docs/deployment.Deploy SIE to our Alibaba Cloud account with the superlinked/sie/alicloud Terraform module. Docs: superlinked.com/docs/deployment.

SIE Local

The open-source server on your own machine.

  • Qwen3.8-27B on one NVIDIA H100, 80 GB
  • Six 61K-token agent sessions per H100
  • Process your data fully offline
  • Same OpenAI-compatible API as Serverless
docker run --gpus all -p 127.0.0.1:8080:8080 ghcr.io/superlinked/sie-server:latest-cuda13-sglang-cu130 serve --host 0.0.0.0 --port 8080 --models-dir /app/models --preload Qwen/Qwen3.8-27B-FP8:h100-256k-batch

Switch to SIE in 5 minutes

{
  "$schema": "https://opencode.ai/config.json",
  "model": "anthropic/claude-sonnet-5-5"
}
from openai import OpenAI
client = OpenAI()
reply = client.chat.completions.create(
    model="gpt-6-luna",
    reasoning_effort="none",
    messages=messages,
    tools=tools,
)
calls = reply.choices[0].message.tool_calls
SIE
{
  "$schema": "https://opencode.ai/config.json",
  "model": "sie/zai-org/GLM-5.3-Flash:relay",
  "provider": {
    "sie": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "SIE",
      "options": {
        "baseURL": "https://api.superlinked.com/v1",
        "apiKey": "{env:SIE_API_KEY}",
        "timeout": 1800000,
        "headerTimeout": 600000,
        "chunkTimeout": 600000
      },
      "models": {
        "zai-org/GLM-5.3-Flash:relay": {
          "name": "GLM-5.3-Flash",
          "limit": {
            "context": 131072,
            "output": 32768
          },
          "tool_call": true,
          "reasoning": true,
          "interleaved": {
            "field": "reasoning_content"
          },
          "options": {
            "include_reasoning": true
          }
        }
      }
    }
  }
}
from openai import OpenAI
client = OpenAI(base_url="https://api.superlinked.com/v1", api_key="sk-sie-…")
reply = client.chat.completions.create(
    model="zai-org/GLM-5.3-Flash:relay",
    messages=messages,
    tools=tools,
)
calls = reply.choices[0].message.tool_calls

Talk to us

Tell us about your use case and we'll get back to you shortly.

Apply for an inference grant

Free capacity on our hosted cluster for selected projects. Tell us what you run and we reply by email.