Groq's LPU hardware delivers unmatched inference speed. Llama-3.3-70b hits up to 300+ tok/s. Best choice when latency matters more than model diversity.

Step models on StepFun's own infrastructure are fast. Best for Chinese/English bilingual tasks.

dola-seed-2.0-lite hits 84 tok/s for coding tasks. Best coding speed per dollar. $9/mo Pro Plan gives access to 9 verified models plus General API.

BytePlus's coding-specific models are purpose-built for code. The dola-seed-2.0-lite hits 84 tok/s, making it both fast and accurate. The $9/mo Pro Plan is excellent value for coding — includes ark-code-latest (auto agent), bytedance-seed-code, and the full General API.

Groq's LPU acceleration makes code generation blazing fast. Free tier is unbeatable for coding throughput.

Ollama offers exclusive coding models like qwen3-coder-next and devstral-2:123b. Unlimited usage on the flat plan.

For sensitive codebases, Chutes offers TEE-enclave execution. Your code stays private even from the provider.

The dola-seed-2.0-pro model is purpose-built for deep reasoning tasks. Shows strong performance on math, logic, and multi-step reasoning benchmarks. $9/mo Pro Plan.

OpenCode offers DeepSeek-V4-Pro at $10/mo flat — the best DeepSeek value. Also offers qwen3-235b reasoning and GLM-5.

ORBIT gives you 2B tokens/month of Claude reasoning for free. Only caveat: only Claude models work reliably.

ZenMux provides the best Gemini 2.5 Pro/Flash access with vision capabilities. 700 free requests/day. Gemini's 1M context window is unmatched for vision tasks.

StepFun's step-3-vl is specifically trained for Chinese/English OCR and vision tasks. Fast and accurate at just $9/mo.

Two vision models for $10/mo flat. Good value with both GLM-5-vision and minimax-vl-01 included.

Ollama offers gemini-3-flash-preview as an exclusive model, plus unlimited usage makes it great for bulk vision tasks.

Groq offers Whisper-large-v3 and Whisper-large-v3-turbo for speech-to-text completely free. LPU acceleration makes it the fastest STT option available. Two STT models at no cost.

StepFun uniquely offers both step-asr (STT) and step-tts (TTS) under one $9/mo plan. Best value if you need both speech input and output.

BytePlus offers Seed Speech TTS and step-asr via the $9/mo Pro Plan. Also includes Seedream image gen, Seedance video gen, and OmniHuman digital human on the same plan.

Infermatic offers Kokoro-82M TTS engine as part of its $20/mo plan, alongside 19 LLM models.

Chutes provides dedicated image generation endpoints. Working models include FLUX.1-schnell (fast), JuggernautXL (cinematic), DreamShaper-XL (artistic), and hunyuan-image-3 (photorealistic). $20/mo PRO plan.

BytePlus offers Seedream-5.0-lite for 4K image generation as part of the $9/mo Pro Plan. Also includes Seedance video gen, making it the best all-in-one creative plan.

Venice offers uncensored image generation. No content filters on prompts or outputs. Ideal for creative freedom.

ArliAI offers image generation API alongside their 54 derestricted LLM models. From $10/mo.

Chutes offers TEE-enclave execution for select models. Your prompts and responses are encrypted in transit and processed inside a hardware-based trusted execution environment. Even Chutes cannot read your data. 20 TEE-enabled models including Qwen, GLM, Kimi, DeepSeek variants.

Venice doesn't log prompts or filter content. No data retention, no surveillance. Best for uncensored workflow. All 75 models have zero logging.

Venice is built from the ground up for uncensored inference. All 75 models have no content safety filters. Claude, Kimi, Grok — all derestricted. No prompt logging, no data retention. Includes grok-4 at 102 tok/s (exclusive).

ArliAI offers derestricted versions of popular models. 54 models with content filters removed. From $10/mo.

Groq offers 16 models completely free with no rate limits on normal usage. LPU hardware means 212+ tok/s average speed. Best free tier in the market by far. Includes both Whisper STT models.

Ollama offers 38+ models including exclusive options like nemotron-3-super, minimax-m2.1, and qwen3-coder-next. Flat rate unlimited plan. Slowest models are still usable.

OpenRouter provides 33 free models (suffix :free). Rate limited to 3-5 req/min per model, but huge variety including Claude, GPT-4, Gemini, and more.

2 billion tokens per month of Claude access for free. Massive allocation but limited to Claude models.

700 free requests per day with good Gemini coverage. Best for moderate usage of Gemini models.

OpenCode gives you access to GLM-5, DeepSeek-V4, MiniMax, and more — all for a flat $10/month. No usage caps, no rate limits. Best per-dollar value by far.

For $9/mo, BytePlus gives you 9 verified models plus General API access to Doubao, DeepSeek-V3.2, Seedream image gen (4K), Seedance video gen, Seed Speech TTS, and OmniHuman digital human. The best all-in-one creative plan.

If you don't need premium models, Groq's free tier with 16 models at 212+ tok/s is unbeatable for cost-conscious usage.

Gemini 2.5 Pro/Flash on ZenMux offer up to 1M token context windows — the largest available. 700 free requests/day. Perfect for document analysis, long-form content, and large codebases.

Claude models on ORBIT support 200K token context. Free with 2B tokens/month. Best free long-context option.

Most Groq models support 128K context. Free and fast — best for moderate-length documents.

BytePlus General API includes Doubao-Pro-128K and Doubao-Lite-128K for long-context tasks. $9/mo Pro Plan.

Chutes offers Qwen3-Embedding-8B inside a TEE enclave for maximum embedding privacy. Your text stays encrypted even from the provider. $20/mo PRO plan.

Infermatic offers the multilingual-e5-base embedding model (768 dimensions) as part of its $20/mo plan alongside 19 LLM models.