Models

Helmcode serves nine open-weight models behind one OpenAI-compatible API at https://api.helmcode.com/v1. Every model works the same way: only the model id changes. Use the id exactly as shown. For limits and quotas, see Rate limits.

Language models

deepseek-v4-flash sota flagship 284B MoE · 21B active · FP8 · 1M context

Flagship for complex, agentic work: long documents, deep codebases, and multi-step tool use.

Capabilities: Tool calling · reasoning · streaming (SSE)
Best for: Reasoning, agents, long-context analysis

qwen3.6 35B MoE · 3B active · FP8 · 256K context

One model, three input types — image, audio, text — and it returns text.

Capabilities: Vision · audio · tool calling · reasoning (max_tokens ≥ 300)
Best for: Multimodal chat, document + media understanding

gemma4 26B MoE · 4B active · FP8 · 256K context

Built for throughput: speculative decoding delivers ~2× tokens per second.

Capabilities: Tool calling (XML) · reasoning · vision · streaming
Best for: High-volume RAG, classification, code completion

Embeddings & reranking

Qw qwen3-embedding 8B · 4096 dimensions · Float32 · MMTEB 70.58

Multilingual semantic embeddings for retrieval and search.

Capabilities: 100+ languages · 60 rpm · batch 32
Best for: RAG retrieval, semantic search, classification

Re rerank Qwen3-Reranker-8B · BF16 · /v1/rerank

Cross-lingual reranking: reorders retrieved passages by actual relevance.

Capabilities: 100+ languages · 1000 rpm
Best for: The middle step of RAG (embedding → rerank → LLM)

Speech

Ko kokoro 82M params · <1s latency · 67 voices

Real-time text-to-speech with sub-second latency.

Capabilities: Voices incl. ef_dora (ES), em_alex (ES), af_heart (EN) · 15 rpm
Best for: Live agents, IVR, voice UIs

whisper-large-v3 99+ languages · 3.2% WER (Spanish)

Speech-to-text with automatic language detection.

Capabilities: 25MB / ~2 min per request · 10 rpm · OGG/Opus, MP3
Best for: Transcription, meeting notes, voice search

On Dedicated and On-premise plans you can also run custom or fine-tuned models on hardware reserved for you. See Examples for how to call each family.