Docscoding planAvailable Models

Available Models

Browse LLMs available through the coding plan API, organized by plan tier.

Overview

The coding plan provides access to a curated set of LLMs for chat, code generation, and completion. Models are organized by plan tier — each tier unlocks access to the full model pool. Use the API to list currently available models.

Free plan

Model IDTierContext WindowQuantization
deepseek-v4-flash-0731Tier 11M tokensFP8

Go plan

Model IDTierContext WindowQuantization
deepseek-v4-flash-0731Tier 11M tokensFP8
google/gemma-3-27b-itTier 1128k tokensFP8
gpt-oss-20bTier 1128k tokensMXFP4
NousResearch/Hermes-4-70BTier 1128k tokensFP8
nvidia/Nemotron-3-Nano-OmniTier 1128k tokensFP8
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3BTier 1128k tokensFP8
gpt-oss-120bTier 2128k tokensMXFP4

Pro plan

Model IDTierContext WindowQuantization
deepseek-v4-flash-0731Tier 11M tokensFP8
gpt-oss-20bTier 1128k tokensMXFP4
NousResearch/Hermes-4-70BTier 1128k tokensFP8
nvidia/Nemotron-3-Nano-OmniTier 1128k tokensFP8
nvidia/NVIDIA-Nemotron-3-Nano-30B-A3BTier 1128k tokensFP8
gpt-oss-120bTier 2128k tokensMXFP4
deepseek-v4-proTier 31M tokensFP8
glm-5.2Tier 31M tokensFP8
kimi-2.7Tier 31M tokensFP8
minmax-m3Tier 31M tokensFP8

Model recommendations

We recommend using deepseek-v4-flash-0731 and Tier 1 models for the best balance of performance, cost, and efficiency.

Tier 2 models consume approximately 2–4× more usage than Tier 1 models.

Tier 3 models consume approximately 8–10× more usage than Tier 1 models.

To ensure the best user experience, we continuously update the models available in each tier based on performance, reliability, availability, and the latest model releases. As newer and better-performing models become available, they may be added to or replace existing models within the respective tiers.

Choosing Tier 1 models whenever possible helps maximize your available usage while maintaining excellent performance for most workloads.

Model selection

When sending a completion request, specify the model in the request body. Use "auto" to let the smart router pick the best model for your prompt.