Skip to main content
Alvin
Alvin
Docs
Supported Models

Supported Models

Models available through the Elyxir API.

The Elyxir API provides access to hundreds of AI models from major providers. This reference lists the currently active models with Elyxir pricing.

OpenAI Models

GPT-5 Series

ModelContextBest For
gpt-5.5128KCurrent flagship
gpt-5.5-pro128KMaximum reasoning depth
gpt-5.5-instant128KSpeed-optimized flagship
gpt-5.4128KMid-tier flagship
gpt-5.2128KPrevious flagship
gpt-5.1128KBudget tier
gpt-5-mini128KBalanced, cost-effective
gpt-5-nano128KFast, budget-friendly

GPT-4.1 Series

ModelContextBest For
gpt-4.11MLong context, vision
gpt-4.1-mini1MEfficient, vision
gpt-4.1-nano1MUltra-fast, cheap

GPT-4 Series

ModelContextBest For
gpt-4o128KGeneral, Vision
gpt-4o-mini128KCost-effective

Embeddings

ModelDimensionsMax Input
text-embedding-3-large30728191
text-embedding-3-small15368191

Anthropic Models

Claude 4.8 Series (Latest)

ModelContextBest For
claude-opus-4.8200KCurrent flagship, agentic coding

Claude 4.7 Series

ModelContextBest For
claude-opus-4.7200KComplex agentic tasks

Claude 4.6 Series

ModelContextBest For
claude-sonnet-4.6200KCoding, Agents
claude-opus-4.6200KMost capable

Claude 4.5 Series

ModelContextBest For
claude-sonnet-4.5200KCoding, Agents
claude-opus-4.5200KMost capable
claude-haiku-4.5200KFast, affordable

Google Models

Gemini 3.5 Series (Latest)

ModelContextBest For
gemini-3.5-flash1MFast, beats 3.1 Pro on coding

Gemini 3.1 Series

ModelContextBest For
gemini-3.1-pro-preview1MMost capable, reasoning

Gemini 3 Series

ModelContextBest For
gemini-3-pro-preview1MAdvanced reasoning
gemini-3-flash1MFast, efficient

Gemini 2.5 Series

ModelContextBest For
gemini-2.5-pro1MBalanced, long context
gemini-2.5-flash1MFast
gemini-2.5-flash-lite1MUltra-fast, low-cost

xAI Models

Grok 4.3 Series (Latest)

ModelContextBest For
grok-4.31MCurrent flagship, cost-efficient

Grok 4 Series

ModelContextBest For
grok-4.61MHigh-capability reasoning
grok-4.51MPrevious-generation reasoning
grok-4.201MAgentic, multi-step work
grok-4.20-multi-agent1MParallel multi-agent runs
grok-4.20-non-reasoning1MFast, no reasoning pass
grok-build-0.11MCode generation

Mistral AI Models

ModelContextBest For
mistral-large-3256KMost capable
mistral-medium-3.1128KBalanced
mistral-small-4256KMultimodal, agentic coding
mistral-small-3.2128KFast, efficient
codestral-22b256KCode generation

DeepSeek Models

ModelContextBest For
deepseek-v4-pro128KFlagship reasoning and coding
deepseek-v4-flash128KFast, cost-effective V4
deepseek-v3.2128KGeneral purpose
deepseek-v3.1128KFast, versatile

Cohere Models

ModelContextBest For
command-a-plus256KLatest flagship, vision
command-a256KEnterprise RAG

Perplexity Models

ModelContextBest For
sonar-deep-research128KDeep research with citations
sonar-reasoning-pro128KReasoning with web search
perplexity-sonar-large128KGeneral search
perplexity-sonar128KFast search

Meta Models

Llama 4 Series

ModelContextBest For
llama4-scout320KEfficient, vision
llama4-maverick1MComplex tasks, vision

Llama 3.3 Series

ModelContextBest For
llama-3.3-70b128KHigh quality

Open-Weight Models (GLM, Kimi, MiniMax, Qwen)

GLM Series (Zhipu AI)

ModelContextBest For
glm-5.1202KCoding, agentic workflows
glm-5202KGeneral purpose
glm-5v-turbo203KNative multimodal (text + image)
glm-4.7128KBalanced
glm-4.6202KCost-efficient

Kimi Series (Moonshot AI)

ModelContextBest For
kimi-k2.6256KMultimodal, long-horizon coding
kimi-k2.5256KAgentic, top-ranked coding
kimi-k2128KAgentic workflows
kimi-k2-thinking128KExtended reasoning

MiniMax Series

ModelContextBest For
minimax-m31MMultimodal flagship
minimax-m2.7200KBalanced MoE
minimax-m2.5196KCost-effective

Qwen Series (Alibaba)

ModelContextBest For
qwen3.7-max1MLong-horizon agent tasks
qwen3.5256KLarge-scale multilingual
qwen3-coder-next256KCode generation
qwen3-235b-thinking128KExtended reasoning
qwen3-32b40KMid-size, versatile

NVIDIA Nemotron Series

ModelContextBest For
nemotron-3-super-120b1MFlagship agentic reasoning
nemotron-3-nano-30b256KFast, low-cost reasoning

Capability-Specific Models

Agentic Coding

ModelContextBest For
grok-build-0.1256KxAI coding model (public beta)
gpt-5.3-codex128KOpenAI Responses API coding

Image Generation

ModelBest For
gpt-image-2Next-gen OpenAI image generation
gpt-image-1.5Fast OpenAI image generation
nano-bananaGemini 2.5 Flash Image, precise text
nano-banana-proGemini 3 Pro Image, studio quality
imagen-4Google Imagen 4 Standard
imagen-4-fast-tgImagen 4 fast tier
imagen-4-ultra-tgImagen 4 ultra tier
flux-2-proFLUX.2 [pro], SOTA open image
flux-2-maxFLUX.2 max tier
flux-2-flexFLUX.2 flex tier
flux-devFLUX.1 dev, high quality
flux-schnellFLUX.1 schnell, ultra-fast
qwen-image-2.0Alibaba Qwen-Image 2.0

Video Generation

ModelBest For
sora-2OpenAI Sora 2
sora-2-proOpenAI Sora 2 Pro

The Veo models (veo-2, veo-3, veo-3-fast, veo-3.1, veo-3.1-fast) are no longer available. Google retired Veo 2.0 and 3.0, and the Veo 3.1 tiers are not currently served. Sora is the video generation route.

Audio

ModelBest For
eleven-v3ElevenLabs v3 TTS, natural voice
tts-1-hdOpenAI HD text-to-speech
tts-1OpenAI standard text-to-speech
gpt-4o-mini-ttsOpenAI mini text-to-speech
scribe-v2ElevenLabs speech-to-text (primary)
scribe-v1ElevenLabs speech-to-text (legacy)
gpt-transcribeOpenAI transcription (current)
gpt-live-transcribeLow-latency streaming transcription
gpt-4o-transcribeGPT-4o powered transcription
gpt-4o-mini-transcribeCheaper GPT-4o transcription
whisper-1OpenAI Whisper transcription

whisper-1, gpt-4o-transcribe and gpt-4o-mini-transcribe are scheduled for retirement on 2027-02-26. gpt-transcribe (file transcription) and gpt-live-transcribe (realtime streaming) are the replacements.

Search

ModelBest For
sonar-deep-researchDeep multi-step research
sonar-reasoning-proReasoning with real-time sources
perplexity-sonar-largeWeb-grounded answers
perplexity-sonarFast web search

Model Selection Guide

By Task

TaskRecommended
General chatgpt-5-mini, claude-haiku-4.5
Complex reasoninggpt-5.5, claude-opus-4.8, deepseek-v4-pro
Code generationclaude-sonnet-4.6, grok-build-0.1
Long documentsgpt-4.1, gemini-2.5-pro
Web searchsonar-deep-research, perplexity-sonar-large
Embeddingstext-embedding-3-small
Fast inferencegpt-5-nano, gemini-2.5-flash-lite
Image generationgpt-image-2, nano-banana-pro
Video generationsora-2, sora-2-pro
Text-to-speecheleven-v3, tts-1-hd

Using Model IDs

Use the full model ID in API requests:

response = client.chat.completions.create(
    model="gpt-4o-mini",  # Model ID
    messages=[...]
)

Use the ID exactly as listed above — the gateway matches it literally. Two things that look reasonable but fail:

  • A provider prefix is not accepted. anthropic/claude-sonnet-4.5 returns a model-not-found; the ID is claude-sonnet-4.5. (The only IDs containing a slash are the self-hosted ollama/* models, where the slash is part of the name itself.)
  • Versions use dots, not dashes. claude-sonnet-4.6 works; claude-sonnet-4-6 does not.

Model Availability

Model availability may vary. Check the Model Hub for current availability and pricing.

Related

Build a ChatbotProviders
Supported Models | Alvin