Baltor Get started

Models

Phi 4

By Microsoft. 14.7 billion parameters, a context window of 16,384 tokens and the licence mit.

Facts

Released
2024-12-11 the day the repository was first published on Hugging Face Hugging Face, read
Licence
mit the licence the model card declares Hugging Face, read
Open weights
Yes the weights are published in this Hugging Face repository Hugging Face, read
Parameters
14.7 billion counted from the safetensors weight files Hugging Face, read
Active parameters
Unknown
Knowledge cutoff
2024-06-30 the knowledge cutoff OpenRouter lists OpenRouter, read
Context window
  • 16,384 tokens the maximum position embeddings in the model configuration model configuration, read
  • 16,384 tokens the context length OpenRouter lists OpenRouter, read
Longest output
  • 14,745 tokens the largest output OpenRouter's first provider allows OpenRouter, read
Tool calling
  • No the chat template in the tokenizer configuration has no place for tool definitions Hugging Face, read
  • No OpenRouter does not list tools among the supported parameters OpenRouter, read
Structured output
  • Yes OpenRouter lists structured outputs among the supported parameters OpenRouter, read
Reasoning controls
Unknown
Inputs and outputs
text in, text out Hugging Face, read
Good for
No source names a use.

Prices

Prices in US dollars per million tokens, as each source lists them
ProviderInputOutputContextAs ofSource
DeepInfra through OpenRouter0.070 USD0.140 USD16,384OpenRouter endpoints, read
Kilo Gateway direct0.070 USD0.140 USD16,384 older than 30 days; check the providermodels.dev, read
OpenRouter direct0.070 USD0.140 USD16,384 older than 30 days; check the providermodels.dev, read

Run it on your own hardware

Weights are the sizes of the files a source lists, or an estimate from the parameter count where none does. Memory is an estimate: weights plus KV cache plus 512 MiB and 5 percent of the weights for runtime buffers. Check it against your hardware.

QuantizationWeightsMemory at 8,192 tokensMemory at 32,768 tokens
Q2_K5.2 GiB7.5 GiBBeyond its context
Q3_K_S6.1 GiB8.4 GiBBeyond its context
Q3_K_M6.9 GiB9.3 GiBBeyond its context
Q3_K_L7.4 GiB9.8 GiBBeyond its context
Q4_K_S7.9 GiB10.3 GiBBeyond its context
Q4_K_M8.4 GiB10.9 GiBBeyond its context
Q5_K_S9.5 GiB12.0 GiBBeyond its context
Q5_K_M9.9 GiB12.4 GiBBeyond its context
Q6_K11.2 GiB13.8 GiBBeyond its context
Q8_014.5 GiB17.3 GiBBeyond its context

Start it with a local runtime

  • Ollama
    ollama run hf.co/MaziyarPanahi/phi-4-GGUF:Q4_K_S
  • LM Studio
    lms get MaziyarPanahi/phi-4-GGUF
lms server start
  • llama.cpp server
    llama-server -hf MaziyarPanahi/phi-4-GGUF:Q4_K_S --jinja
  • vLLM
    vllm serve microsoft/phi-4
  • SGLang
    sglang serve --model-path microsoft/phi-4 --port 30000

Use it from your harness

Set up for Kilo Gateway with the model microsoft/phi-4. Each endpoint page has the same setup for its own address.

OpenCode

Put this in opencode.json in your project folder:

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "kilo": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Kilo Gateway",
      "options": {
        "baseURL": "https://api.kilo.ai/api/gateway",
        "apiKey": "{env:KILO_API_KEY}"
      },
      "models": {
        "microsoft/phi-4": {
          "name": "Phi 4",
          "limit": {
            "context": 16384,
            "output": 14745
          }
        }
      }
    }
  }
}
  • OpenCode reads any OpenAI-compatible address through the @ai-sdk/openai-compatible package, and an address that speaks the Responses API through @ai-sdk/openai.

From OpenCode documentation, read .

Pi

Put this in ~/.pi/agent/models.json:

{
  "providers": {
    "kilo": {
      "baseUrl": "https://api.kilo.ai/api/gateway",
      "api": "openai-completions",
      "apiKey": "$KILO_API_KEY",
      "models": [
        {
          "id": "microsoft/phi-4"
        }
      ]
    }
  }
}
  • The apiKey field can name an environment variable as $NAME.

From Pi documentation, read .

Codex

Codex speaks only the Responses API, and Kilo Gateway documents no Responses address. A gateway that offers one can sit in between.

  • Codex speaks the Responses API only: responses is the one supported wire API of a custom provider. Ollama and LM Studio are built in and start with --oss.

From Codex documentation, read .

Claude Code

Claude Code sends Anthropic Messages requests, and Kilo Gateway documents no such address. A gateway that translates to that API can sit in between.

  • Claude Code sends Anthropic Messages requests to ANTHROPIC_BASE_URL. Anthropic says it does not support routing Claude Code to models other than Claude through any gateway, so some features may not work with another model.

From Claude Code documentation, read .

Other ways to reach it

  • OpenRouter, model microsoft/phi-4
  • Ollama, model hf.co/MaziyarPanahi/phi-4-GGUF:Q4_K_S

Published results

These are results other people published. Baltor did not run them and does not rank models by them.