Free LLM models
via API.

Last scan discovering 20 free Models from 13 Providers via 5 Aggregators for you.

Aggregators

Providers

Upstage

Solar Mini 4

  1. 🏆 Nous Portal
CONTEXT524.29K
MAX OUTPUT131.07K
Reasoning Text
Official model page

Upstage's compact mixture-of-experts language model for retrieval, structured outputs and tool-assisted workflows. It has 35 billion total parameters and activates 3 billion per token, with optional reasoning.

Details & thinking

The creator describes limited-time free Hermes Agent access starting October 5, 2026. Nous currently lists Solar Mini 4 as free; continued free availability is not established.

Limits above: Nous Portal. Available gateway routes are listed below.

The supplied Nous route upstage/solar-mini4:free and the official Nous listing identify Solar Mini 4 by Upstage, matching the creator's announcement and solar-mini4 Chat API alias. The gateway does not establish a pinned checkpoint revision. Solar Mini 4 is distinct from the known Solar Pro 4 entry; no known sibling requires identity merging or display grouping.

Nous Portal
Info
Thinking
Supported
Effort levels
max · xhigh · high · medium · low · minimal · none
Control
Can be toggled
Default
Disabled
Default effort
medium
Model ID · Nous Portal APIupstage/solar-mini4:free

Model sources

Try this model
Nous Portal · Not yet tested

InclusionAI

Ling 3.1 Flash

  1. 🏆 Kilo
  2. 🏆 OpenRouter
CONTEXT262.14K
MAX OUTPUT32.77K
Reasoning Text

InclusionAI's hybrid reasoning language model, offered by Kilo with text input, tool calling and reasoning support.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo's exact listing at https://kilo.ai/models/inclusionai-ling-3-1-flash identifies inclusionai/ling-3.1-flash and its creator. This matches OpenRouter's exact-version catalog entry at https://openrouter.ai/api/v1/models. OpenCode explicitly maps its free alias to this namespaced model in https://raw.githubusercontent.com/anomalyco/models.dev/dev/providers/opencode/models/ling-3.1-flash-free.toml. Gateway configurations and pinned weight revisions are not assumed identical.

Kilo
Info
Thinking
Supported
Model ID · Kilo APIinclusionai/ling-3.1-flash
OpenRouter
Info
Thinking
Supported
Control
Can be toggled
Default
Enabled
Model ID · OpenRouter APIinclusionai/ling-3.1-flash

Model sources

Try this model
Kilo · Last tested

Apodex

Apodex 1.1 Mini

  1. 🏆 Kilo
  2. 🏆 OpenRouter
CONTEXT262.14K
MAX OUTPUT235.93K
Reasoning Open weights Text
Official model page

Apodex's compact reasoning model for research, coding and tool-assisted workflows. The supplied Kilo catalog lists text input and output.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

The supplied Kilo catalog snapshot identifies apodex/apodex-1.1-mini:free, matching the exact namespaced version in https://openrouter.ai/api/v1/models and the creator's repository at https://huggingface.co/apodex/Apodex-1.1-mini. The shared identity denotes Apodex 1.1 Mini; it does not establish identical gateway configurations or pinned weight revisions.

Kilo
Info
Thinking
Supported
Model ID · Kilo APIapodex/apodex-1.1-mini:free
OpenRouter
Info
Thinking
Supported
Control
Can be toggled
Model ID · OpenRouter APIapodex/apodex-1.1-mini:free

Model sources

Try this model
Kilo · Last tested

Meituan

LongCat 2.5 Preview

  1. 🏆 Nous Portal
CONTEXT1.05M
MAX OUTPUT131.07K
Reasoning TextImage
Official model page

Meituan's multimodal coding and agent model, with image understanding and long-context reasoning. Its thinking mode can be switched on or off; named effort levels are not documented.

Details & thinking

Limits above: Nous Portal. Available gateway routes are listed below.

Reviewed as the same advertised Meituan LongCat 2.5 Preview model across Nous Portal and OpenCode Zen, not LongCat 2.0. Gateway limits and access conditions remain separate.

Nous Portal
Info
Thinking
Supported
Control
Can be toggled
Default
Enabled
Model ID · Nous Portal APImeituan/longcat-2.5-preview:free

Model sources

Try this model
Nous Portal · Last tested

Anonymous · stealth preview

Space Bunny

  1. 🏆 OpenCode Zen
CONTEXT1.05M
MAX OUTPUT524.29K
TextImageVideo

An anonymous preview model advertised for coding, reasoning and multimodal input. Its developer and underlying model identity have not been disclosed.

Details & thinking

Limits above: OpenCode Zen. Available gateway routes are listed below.

OpenCode lists space-bunny-free as a stealth chat model at https://opencode.ai/docs/zen/. This does not establish identical weights with https://openrouter.ai/stealth/space-bunny-alpha. Replaced the previous shared alias-family key with a route-specific identity because the previous identity note itself acknowledged that exact equivalence was unverified.

OpenCode Zen
Info
Thinking
Not advertised by this API
Model ID · OpenCode Zen APIspace-bunny-free

Model sources

Try this model
OpenCode Zen · Last tested

SDAIA

ALLaM-2-7b

  • Groq
CONTEXT4.1K
MAX OUTPUT4.1K
Text
Official model page

SDAIA's 7-billion-parameter instruction-tuned model for Arabic and English. Trained from scratch with staged English and Arabic-English pretraining, it supports bilingual conversations, text generation and summarization.

Details & thinking

Limits above: Groq. Available gateway routes are listed below.

Groq
Info
Thinking
Not advertised by this API
Model ID · Groq APIallam-2-7b

Model sources

Try this model
Groq · Last tested

Cohere

North Mini Code

  • Kilo
  • OpenRouter
CONTEXT256K
MAX OUTPUT64K
Reasoning Open weights Text
Official model page

Cohere's compact mixture-of-experts model focused on agentic coding. It targets code changes and tool-driven software development.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APIcohere/north-mini-code:free
OpenRouter
Info
Thinking
Supported
Control
Can be toggled
Model ID · OpenRouter APIcohere/north-mini-code:free

Model sources

Try this model
Kilo · Last tested

Dots Studio

Dots3-Note Preview

  • Kilo
  • OpenRouter
CONTEXT512K
MAX OUTPUT460.8K
Reasoning Open weights TextImage
Official model page

A preview of Dots Studio's lighter Dots 3 mixture-of-experts model. It supports long-context work and configurable reasoning.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APIdots-studio/dots-3-note-preview:free
OpenRouter
Info
Thinking
Supported
Control
Can be toggled
Model ID · OpenRouter APIdots-studio/dots-3-note-preview:free

Model sources

Try this model
Kilo · Last tested

OpenAI

GPT OSS 120B

  • Groq
CONTEXT131.07K
MAX OUTPUT65.54K
Open weights Text
Official model page

OpenAI's larger open-weight reasoning model, served here by Groq. It offers adjustable reasoning effort for text and tool-based tasks.

Details & thinking

Limits above: Groq. Available gateway routes are listed below.

Groq
Info
Thinking
Not advertised by this API
Model ID · Groq APIopenai/gpt-oss-120b

Model sources

Try this model
Groq · Last tested

OpenAI

GPT OSS 20B

  • Groq
CONTEXT131.07K
MAX OUTPUT65.54K
Open weights Text
Official model page

The smaller open-weight GPT-OSS reasoning model, served here by Groq. It supports low, medium and high reasoning effort.

Details & thinking

Limits above: Groq. Available gateway routes are listed below.

Groq
Info
Thinking
Not advertised by this API
Model ID · Groq APIopenai/gpt-oss-20b

Model sources

Try this model
Groq · Last tested

InclusionAI

Ling 3.0 Flash Sante

  • Kilo
  • Nous Portal
  • OpenRouter
CONTEXT262.14K
MAX OUTPUT32.77K
Reasoning Text
Official announcement

A health- and medicine-focused Ling 3.0 Flash model from InclusionAI. Domain specialization does not make its responses medical advice.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APIinclusionai/ling-3.0-flash-sante:free
Nous Portal
Info
Thinking
Supported
Control
Can be toggled
Default
Enabled
Model ID · Nous Portal APIinclusionai/ling-3.0-flash-sante:free
OpenRouter
Info
Thinking
Supported
Control
Can be toggled
Default
Enabled
Model ID · OpenRouter APIinclusionai/ling-3.0-flash-sante:free

Model sources

Try this model
Kilo · Last tested

Liquid AI

LFM2.5-2.6B

  • Kilo
  • OpenRouter
CONTEXT65.54K
MAX OUTPUT8.19K
Reasoning Open weights Text
Official model page

Liquid AI's compact reasoning model for extraction, retrieval and agent workflows. Its model guidance does not position it as an agentic coding specialist.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APIliquid/lfm-2.5-2.6b:free
OpenRouter
Info
Thinking
Supported
Control
Always enabled
Model ID · OpenRouter APIliquid/lfm-2.5-2.6b:free

Model sources

Try this model
Kilo · Last tested

NVIDIA

Nemotron 3 Nano Omni

  • OpenRouter
CONTEXT256K
MAX OUTPUT65.54K
Reasoning Open weights TextAudioImageVideo
Official model page

A multimodal NVIDIA model that can work with text, images, audio and video. Designed for perception and context processing in agent workflows.

Details & thinking

Limits above: OpenRouter. Available gateway routes are listed below.

OpenRouter
Info
Thinking
Supported
Control
Can be toggled
Default
Enabled
Token budget
Supported
Model ID · OpenRouter APInvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free

Model sources

Try this model
OpenRouter · Not yet tested

NVIDIA

Nemotron 3 Super

  • Kilo
  • OpenRouter
CONTEXT262.14K
MAX OUTPUT235.93K
Reasoning Open weights Text
Official model page

NVIDIA's hybrid mixture-of-experts reasoning model for multi-agent workflows. Only a subset of its total parameters is active per token.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APInvidia/nemotron-3-super-120b-a12b:free
OpenRouter
Info
Thinking
Supported
Effort levels
medium · low
Control
Can be toggled
Default
Enabled
Default effort
medium
Token budget
Supported
Model ID · OpenRouter APInvidia/nemotron-3-super-120b-a12b:free

Model sources

Try this model
Kilo · Last tested

NVIDIA

Nemotron 3 Ultra

  • Kilo
  • OpenRouter
CONTEXT1M
MAX OUTPUT65.54K
Reasoning Open weights Text
Official model page

NVIDIA's larger Nemotron 3 model for reasoning and agent orchestration. It combines a long context window with a sparse hybrid architecture.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APInvidia/nemotron-3-ultra-550b-a55b:free
OpenRouter
Info
Thinking
Supported
Effort levels
high · medium
Control
Can be toggled
Default
Enabled
Default effort
high
Token budget
Supported
Model ID · OpenRouter APInvidia/nemotron-3-ultra-550b-a55b:free

Model sources

Try this model
Kilo · Last tested

NVIDIA

Nemotron 3.5 Lightning

  • Kilo
  • OpenRouter
CONTEXT1M
MAX OUTPUT65.54K
Reasoning Open weights Text
Official model page

NVIDIA's lightweight mixture-of-experts model for responsive agent workflows. It balances a small active parameter count with long-context processing.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APInvidia/nemotron-3.5-lightning:free
OpenRouter
Info
Thinking
Supported
Control
Can be toggled
Model ID · OpenRouter APInvidia/nemotron-3.5-lightning:free

Model sources

Try this model
Kilo · Last tested

Poolside

Laguna S 2.1

  • Kilo
  • Nous Portal
CONTEXT262.14K
MAX OUTPUT32.77K
Reasoning Open weights Text
Official model page

Poolside's coding-agent model for tool-driven software engineering. Gateway-specific context and output limits are listed separately.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APIpoolside/laguna-s-2.1:free
Nous Portal
Info
Thinking
Supported
Control
Can be toggled
Default
Enabled
Model ID · Nous Portal APIpoolside/laguna-s-2.1:free
Max output here
131,072

Model sources

Try this model
Kilo · Last tested

Poolside

Laguna XS 2.1

  • Nous Portal
  • OpenRouter
CONTEXT262.14K
MAX OUTPUT32.77K
Reasoning Open weights Text
Official model page

A smaller coding-agent model in Poolside's Laguna family. It is designed for software development workflows with a relatively small active parameter count.

Details & thinking

Limits above: Nous Portal. Available gateway routes are listed below.

Nous Portal
Info
Thinking
Supported
Control
Can be toggled
Default
Enabled
Model ID · Nous Portal APIpoolside/laguna-xs-2.1:free
OpenRouter
Info
Thinking
Supported
Control
Can be toggled
Default
Enabled
Model ID · OpenRouter APIpoolside/laguna-xs-2.1:free

Model sources

Try this model
Nous Portal · Last tested

Alibaba

Qwen3.8 27B

  • Groq
CONTEXT131.07K
MAX OUTPUT16.38K
Open weights TextImage
Official model page

A dense Qwen vision-language model for coding, visual analysis and agent tasks. Thinking controls and context limits vary by gateway.

Details & thinking

Limits above: Groq. Available gateway routes are listed below.

Groq
Info
Thinking
Not advertised by this API
Model ID · Groq APIqwen/qwen3.8-27b

Model sources

Try this model
Groq · Last tested

StepFun

Step 3.7 Flash

  • Kilo
  • Nous Portal
CONTEXT262.14K
MAX OUTPUT262.14K
Reasoning Open weights TextImage
Official model page

StepFun's multimodal mixture-of-experts model with image and video understanding. It combines reasoning and tool use in a relatively efficient architecture.

Details & thinking

Limits above: Kilo. Available gateway routes are listed below.

Kilo
Info
Thinking
Supported
Model ID · Kilo APIstepfun/step-3.7-flash:free
Nous Portal
Info
Thinking
Supported
Effort levels
high · medium · low
Control
Always enabled
Default effort
medium
Model ID · Nous Portal APIstepfun/step-3.7-flash:free
Max output here
256,000

Model sources

Try this model
Kilo · Last tested

Aggregator guide

Groq

Low-latency inference through the GroqCloud API.

Create account API documentation

Access & limits

A free developer tier with per-model limits.

API connection

Base URL https://api.groq.com/openai/v1

Use your Groq API key and the Model ID shown under Groq in the model details. Keep the ID exactly as listed.

Do not add a gateway prefix or a local -fast / -think suffix. Thinking settings are listed separately for each model and aggregator.

Sources & references

Back to the directory ↑

Aggregator guide

Kilo

A unified gateway with rotating free model offers.

Create account API documentation

Access & limits

Free availability and upstream data policies vary.

API connection

Base URL https://api.kilo.ai/api/gateway

Use your Kilo API key and the Model ID shown under Kilo in the model details. Keep the ID exactly as listed.

Do not add a gateway prefix or a local -fast / -think suffix. Thinking settings are listed separately for each model and aggregator.

Sources & references

Back to the directory ↑

Aggregator guide

Nous Portal

Nous Research's portal for hosted model access.

Create account API documentation

Access & limits

The Free plan provides free models with standard rate limits and no monthly credits; available routes can change.

API connection

Base URL https://inference-api.nousresearch.com/v1

Use your Nous Portal API key and the Model ID shown under Nous Portal in the model details. Keep the ID exactly as listed.

Do not add a gateway prefix or a local -fast / -think suffix. Thinking settings are listed separately for each model and aggregator.

Sources & references

Back to the directory ↑

Aggregator guide

NVIDIA

Hosted model previews through the NVIDIA API catalog.

Create account API documentation

Access & limits

API trial terms, quotas and model licenses apply.

API connection

Base URL https://integrate.api.nvidia.com/v1

Use your NVIDIA API key and the Model ID shown under NVIDIA in the model details. Keep the ID exactly as listed.

Do not add a gateway prefix or a local -fast / -think suffix. Thinking settings are listed separately for each model and aggregator.

Sources & references

Back to the directory ↑

Aggregator guide

OpenCode Zen

Models curated for coding and agent workflows.

Create account API documentation

Access & limits

⚠️ Some free routes require the OpenCode app or CLI.

Selected models are temporarily free. Direct API endpoints are documented; a particular free route may still restrict access to the OpenCode app. The scan checks each route separately. Direct API availability varies by model; the directory shows each route's result from the latest scan.

API connection

Base URL https://opencode.ai/zen/v1

Use your OpenCode Zen API key and the Model ID shown under OpenCode Zen in the model details. Keep the ID exactly as listed.

Do not add a gateway prefix or a local -fast / -think suffix. Thinking settings are listed separately for each model and aggregator.

Sources & references

Back to the directory ↑

Aggregator guide

OpenRouter

One API, many model creators and inference providers.

Create account API documentation

Access & limits

Free-model quotas and model-specific terms apply.

API connection

Base URL https://openrouter.ai/api/v1

Use your OpenRouter API key and the Model ID shown under OpenRouter in the model details. Keep the ID exactly as listed.

Do not add a gateway prefix or a local -fast / -think suffix. Thinking settings are listed separately for each model and aggregator.

Sources & references

Back to the directory ↑