Skip to content

FREE_MODELS.md

Latest commit

 

History

History
328 lines (244 loc) · 14.4 KB

FREE_MODELS.md

File metadata and controls

328 lines (244 loc) · 14.4 KB

Free Models Support in Hive-Mind (languages: en • zh • hi • ru)

This document provides comprehensive information about the free models supported by hive-mind when using the --tool agent option.

Last Updated: April 10, 2026 Related:

Available Free Models

Hive-mind supports free models from two providers:

  1. OpenCode Zen - 4 free models with opencode/ prefix
  2. Kilo Gateway - 6 free models with kilo/ prefix (Issue #1282)

OpenCode Zen Free Models

1. opencode/nemotron-3-super-free Default Model

  • Short Alias: nemotron-3-super-free
  • Provider: OpenCode Zen
  • Status: Fully Supported (Default for --tool agent as of Issue #1563)
  • Features: Reasoning, tool calling, hybrid Mamba-Transformer architecture
  • Context Window: ~262,144 tokens
  • Output Limit: 262,144 tokens
  • Cost: Free (no input/output charges)
  • Knowledge Cutoff: January 2025
  • Release Date: March 2026
  • Open Weights: Yes
  • Notes: NVIDIA hybrid Mamba-Transformer MoE, strong reasoning capabilities

2. opencode/minimax-m2.5-free

  • Short Alias: minimax-m2.5-free
  • Provider: OpenCode Zen
  • Status: Fully Supported (Former default, Issues #1391, #1543)
  • Features: Reasoning, tool calling, temperature control
  • Context Window: 204,800 tokens
  • Output Limit: 131,072 tokens
  • Cost: Free (no input/output charges)
  • Knowledge Cutoff: January 2025
  • Release Date: February 2026
  • Open Weights: Yes

3. opencode/gpt-5-nano

  • Short Alias: gpt-5-nano
  • Provider: OpenCode Zen
  • Status: Fully Supported
  • Features: Reasoning, tool calling, structured output, temperature control
  • Context Window: ~400,000 tokens
  • Output Limit: 128,000 tokens
  • Cost: Free (no input/output charges)
  • Knowledge Cutoff: January 2025

4. opencode/big-pickle

  • Short Alias: big-pickle
  • Provider: OpenCode Zen
  • Status: Fully Supported
  • Features: Reasoning, tool calling, temperature control
  • Context Window: ~200,000 tokens
  • Output Limit: 128,000 tokens
  • Cost: Free (no input/output charges)
  • Knowledge Cutoff: January 2025

Discontinued OpenCode Zen Free Models

The following models were previously free but are no longer available:

Model Former Model ID Status
Qwen 3.6 Plus Free opencode/qwen3.6-plus-free Free promotion ended (April 2026) — now requires OpenCode Go subscription. See agent#242
Kimi K2.5 Free opencode/kimi-k2.5-free Removed from OpenCode Zen (March 2026) — see agent#208
Grok Code Fast 1 opencode/grok-code Discontinued January 2026
MiniMax M2.1 Free opencode/minimax-m2.1-free Replaced by opencode/minimax-m2.5-free
GLM 4.7 Free opencode/glm-4.7-free No longer free on OpenCode Zen

Note: See OpenCode Zen Documentation and Agent CLI FREE_MODELS.md for the current list of free models.

Kilo Gateway Free Models

Kilo Gateway provides access to 500+ AI models through an OpenAI-compatible API. The following free models are available without API key configuration.

Note: Kilo-exclusive models (models only available on Kilo Gateway) support short aliases without the kilo/ prefix. For example, you can use glm-5-free instead of kilo/glm-5-free since this model is unique to Kilo.

1. kilo/glm-5-free Recommended for Kilo

  • Model ID: kilo/glm-5-free
  • Short Alias: glm-5-free (Kilo-exclusive model)
  • Provider: Kilo Gateway (Z.AI)
  • Status: Fully Supported (Free for limited time)
  • Features: Deep reasoning, fast inference, bilingual (Chinese/English), tool calling, structured outputs
  • Context Window: 202,752 tokens
  • Output Limit: 131,072 tokens
  • Cost: Free (limited time offer)
  • Release Date: February 11, 2026
  • Special Features: "Matches Opus 4.5 on many tasks" - Kilo Blog

2. kilo/glm-4.5-air-free

  • Model ID: kilo/glm-4.5-air-free
  • Short Alias: glm-4.5-air-free (Kilo-exclusive model)
  • Provider: Kilo Gateway (Z.AI)
  • Status: Fully Supported
  • Features: Agent-centric, lightweight, fast inference
  • Context Window: 131,072 tokens
  • Output Limit: 65,536 tokens
  • Cost: Free

3. kilo/minimax-m2.5-free

  • Model ID: kilo/minimax-m2.5-free
  • Provider: Kilo Gateway (MiniMax)
  • Status: Fully Supported (upgraded from M2.1)
  • Features: Strong general-purpose performance
  • Context Window: 204,800 tokens
  • Output Limit: 131,072 tokens
  • Cost: Free

4. kilo/deepseek-r1-free

  • Model ID: kilo/deepseek-r1-free
  • Short Alias: deepseek-r1-free (Kilo-exclusive model)
  • Provider: Kilo Gateway (DeepSeek)
  • Status: Fully Supported
  • Features: Advanced reasoning, open-source, fully open reasoning tokens
  • Context Window: 163,840 tokens
  • Output Limit: 65,536 tokens
  • Cost: Free

5. kilo/giga-potato-free

  • Model ID: kilo/giga-potato-free
  • Short Alias: giga-potato-free (Kilo-exclusive model)
  • Provider: Kilo Gateway
  • Status: Fully Supported (Evaluation period)
  • Features: General-purpose evaluation model
  • Context Window: 256,000 tokens
  • Output Limit: 131,072 tokens
  • Cost: Free (during evaluation)

6. kilo/trinity-large-preview

  • Model ID: kilo/trinity-large-preview
  • Short Alias: trinity-large-preview (Kilo-exclusive model)
  • Provider: Kilo Gateway (Arcee AI)
  • Status: Fully Supported (Preview)
  • Features: Strong capabilities, preview model
  • Context Window: 131,000 tokens
  • Output Limit: 65,536 tokens
  • Cost: Free (preview)

Discontinued Kilo Gateway Free Models

The following Kilo models were previously the recommended free models but have been updated:

Model Former Model ID Status
GLM 4.7 kilo/glm-4.7-free Replaced by kilo/glm-4.5-air-free
Kimi K2.5 kilo/kimi-k2.5-free Replaced by other Kilo free models
MiniMax M2.1 kilo/minimax-m2.1-free Replaced by kilo/minimax-m2.5-free

Note: See Kilo Free Models Documentation for current availability.

Usage Examples

Command Line Usage

# OpenCode Zen models (short aliases without prefix)
solve https://github.com/owner/repo/issues/123 --tool agent --model nemotron-3-super-free
hive https://github.com/owner/repo --tool agent --model minimax-m2.5-free

# OpenCode Zen models (full model IDs)
solve https://github.com/owner/repo/issues/123 --tool agent --model opencode/nemotron-3-super-free
hive https://github.com/owner/repo --tool agent --model opencode/big-pickle

# Kilo Gateway models (full model IDs)
solve https://github.com/owner/repo/issues/123 --tool agent --model kilo/glm-5-free
hive https://github.com/owner/repo --tool agent --model kilo/deepseek-r1-free

# Kilo-exclusive models (short aliases without kilo/ prefix)
solve https://github.com/owner/repo/issues/123 --tool agent --model glm-5-free
hive https://github.com/owner/repo --tool agent --model deepseek-r1-free

Telegram Bot Usage

# OpenCode Zen models (short aliases)
/solve https://github.com/owner/repo/issues/123 --tool agent --model nemotron-3-super-free
/solve https://github.com/owner/repo/issues/123 --tool agent --model minimax-m2.5-free
/agent https://github.com/owner/repo/issues/123 --model nemotron-3-super-free

# Kilo Gateway models (full model IDs)
/solve https://github.com/owner/repo/issues/123 --tool agent --model kilo/glm-5-free
/hive https://github.com/owner/repo --tool agent --model kilo/glm-4.5-air-free

# Kilo-exclusive models (short aliases without kilo/ prefix)
/solve https://github.com/owner/repo/issues/123 --tool agent --model glm-5-free
/hive https://github.com/owner/repo --tool agent --model glm-4.5-air-free

# Default model (nemotron-3-super-free via OpenCode Zen):
/solve https://github.com/owner/repo/issues/123 --tool agent
/agent https://github.com/owner/repo/issues/123

Direct Agent CLI Usage

# OpenCode Zen models
echo "Your prompt here" | agent --model opencode/nemotron-3-super-free
echo "Your prompt here" | agent --model opencode/minimax-m2.5-free

# Kilo Gateway models
echo "Your prompt here" | agent --model kilo/glm-5-free
echo "Your prompt here" | agent --model kilo/deepseek-r1-free

Model Selection Guide

For Different Use Cases

Flagship Free Models:

  • opencode/nemotron-3-super-free - NVIDIA hybrid Mamba-Transformer, strong reasoning (OpenCode, default)
  • kilo/glm-5-free - Z.AI flagship, matches Opus 4.5 on many tasks (Kilo)

General Purpose & Reasoning:

  • opencode/gpt-5-nano - Strong general reasoning capabilities
  • opencode/big-pickle - Well-balanced performance
  • kilo/minimax-m2.5-free - Strong general-purpose performance
  • kilo/deepseek-r1-free - Advanced reasoning model

For Large Context Tasks:

  • opencode/gpt-5-nano - Very large context (~400,000 tokens)
  • opencode/nemotron-3-super-free - Large context (~262,144 tokens)
  • kilo/giga-potato-free - Large context (256,000 tokens)
  • opencode/minimax-m2.5-free - Large context (204,800 tokens)

Agent-Centric / Coding:

  • kilo/glm-4.5-air-free - Purpose-built for agent-centric applications
  • kilo/deepseek-r1-free - Optimized for reasoning and code synthesis
  • opencode/minimax-m2.5-free - Strong coding performance

Provider Comparison

Feature OpenCode Zen Kilo Gateway
Free Models 4 models 6 models
Default Model nemotron-3-super-free (~262K context) glm-5-free (recommended)
API Format OpenAI-compatible OpenAI-compatible
Free API Key public public
Total Models 50+ 500+
Flagship Free Nemotron 3 Super (~262K context) GLM-5 (limited time)
BYOK Support Yes Yes
New Models Nemotron 3 Super (Issue #1543, #1563) DeepSeek R1, GLM 4.5 Air

Testing and Validation

All free models have been tested and validated for:

  1. Model Configuration: All models are properly configured in src/models/index.mjs
  2. CLI Integration: All models are accepted by both hive-mind and agent CLI
  3. Tool Compatibility: All models are compatible with the --tool agent option
  4. Case Insensitive Usage: Models can be specified in any case (e.g., KILO/GLM-5-FREE)
  5. Alias Support: Short aliases work for all models

Error Handling

If you encounter issues with any of these models:

  1. Check Model Spelling: Ensure exact model name or alias is used
  2. Update Dependencies: Run npm install to ensure latest agent CLI
  3. Check Network: Some models may require internet access for first-time setup
  4. Verify Provider: Ensure correct provider prefix (opencode/ or kilo/)

Related Documentation

Last Updated: April 10, 2026 Hive-Mind Version: 1.48.2 Agent CLI Version: Latest (with free model updates from PR #243)