← Home
LLM Directory
Compare pricing, context windows, and benchmarks across 164 models from 16 providers and 3 hosted platforms.
LLM Providers
Companies that build and host their own foundation models.
OpenAI
22 models
Leading AI lab and API provider. Creators of GPT, o-series reasoning models, and DALL·E.
From $5.00Budget $0.040
Anthropic
9 models
AI safety company. Creators of the Claude model family, emphasizing responsible AI development.
From $5.00Budget $0.625
15 models
Google DeepMind's Gemini model family. 1M+ context windows, multimodal, competitive pricing.
From $2.00Budget $0.020
DeepSeek
7 models
Chinese AI lab offering frontier open-weight models at dramatically low prices.
From $0.003
xAI
5 models
Elon Musk's AI company. Grok models with real-time X (Twitter) data access.
From $3.00Budget $0.200
Meta
10 models
Meta's open-weight Llama model family. Free to download; pricing shown for hosted API partners.
From $3.00Budget $0.050
Mistral
8 models
French AI company offering open-weight and commercial models. Known for efficient architectures.
From $1.36Budget $0.100
Cohere
6 models
Enterprise-focused AI company. Command R models for RAG, search, and business applications.
From $2.50Budget $0.050
Moonshot AI
5 models
Chinese AI company. Creators of Kimi models with strong long-context and coding abilities.
From $0.600Budget $0.150
MiniMax
5 models
Chinese AI company offering text, speech, and video generation models.
From $0.300Budget $0.010
Alibaba Cloud (Qwen)
10 models
Alibaba's Qwen model family. Open-weight and API-available. Strong multilingual support.
From $0.800Budget $0.010
NVIDIA
5 models
NVIDIA NIM inference platform. Hosts and serves popular open-weight models at scale.
From $0.800Budget $0.200
Shanghai AI Laboratory
1 models
Shanghai AI Lab (InternLM) — agentic models with open MIT weights and long context windows.
Aleph Alpha
1 models
German AI lab. Kolibri 1 is an Apache 2.0 English-German mixture-of-experts model. No public token price was on the model card or the October 3, 2026 launch post.
Cloudflare
3 models
Workers AI hosts Clef, Clef-flash, and Clef-Omni. All three are decision models. The published unit is input tokens. No output token price is on the model pages.
Budget $0.090
Perplexity
1 models
pplx-decider-v1-27b is an Apache 2.0 decision checkpoint. The model card publishes no token price and no context length.
Hosted Platforms
Unified APIs that serve models from multiple providers — one key, many models.
Ollama
20 models
Run 100+ open-weight models locally. No API costs. The easiest way to self-host LLMs on Linux, macOS, and Windows.
20 free
OpenRouter
20 models
Unified API gateway to 400+ models from OpenAI, Anthropic, Google, Meta, and more. Pay per token, no subscriptions.
$0.050 – $15.00/1M
Together AI
11 models
Serverless GPU cloud for open-weight models. Fast inference, competitive pricing, 200+ models.
$0.010 – $0.880/1M
Full Comparison
All models sortable by provider, price, and context window. Click a provider name for details.
| Provider | Model | Category | Input / 1M | Output / 1M | Context |
|---|---|---|---|---|---|
| OpenAI | GPT-5.5 OpenAI's widely available 2026 coding and reasoning workhorse for ChatGPT, Codex, and the API. Strong tool use with large context. | Flagship | $5.00 | $30.00 | 1.0M |
GPT-5.6 OpenAI's July 2026 three-tier family (Sol / Terra / Luna) with programmatic tool calling in the Responses API. Sol is the frontier tier for coding, reasoning, and agent workflows. | Flagship | $5.00 | $30.00 | 1.0M | |
GPT-4.1 Long-context coding model with 1M token context window. | Flagship | $2.00 | $8.00 | 1.0M | |
GPT-4.1 Mini Affordable long-context model balancing speed and intelligence. | Mid-range | $0.400 | $1.60 | 1.0M | |
GPT-4.1 Nano Ultra-fast, ultra-cheap model for high-volume tasks. | Budget | $0.100 | $0.400 | 1.0M | |
GPT-4o Flagship multimodal model. Fast, capable, and widely deployed. | Flagship | $2.50 | $10.00 | 128K | |
GPT-4o Mini Fast, affordable small model for everyday tasks. | Budget | $0.150 | $0.600 | 128K | |
o3 Reasoning model for complex STEM, math, and multi-step tasks. | Reasoning | $10.00 | $40.00 | 200K | |
o3 Pro Highest-capability reasoning model for hardest problems. | Reasoning | $20.00 | $80.00 | 200K | |
o3 Deep Research Extended reasoning with web browsing for research tasks. | Reasoning | $10.00 | $40.00 | 200K | |
o4-mini Affordable reasoning model balancing cost and capability. | Reasoning | $1.10 | $4.40 | 200K | |
o4-mini Deep Research Budget research model with web browsing. | Reasoning | $1.10 | $4.40 | 200K | |
GPT-5 Next-generation general-purpose model with agentic capabilities. | Flagship | $1.25 | $10.00 | 128K | |
GPT-5.1 Improved GPT-5 with larger context and better reasoning. | Flagship | $2.00 | $12.00 | 256K | |
GPT-5.2 Latest GPT-5 iteration with strongest coding and math. | Flagship | $2.50 | $15.00 | 256K | |
GPT-5 Mini Fast, affordable variant of GPT-5 for everyday tasks. | Mid-range | $0.250 | $1.50 | 128K | |
GPT-5 Nano Ultra-cheap GPT-5 variant for high-volume classification. | Budget | $0.050 | $0.300 | 128K | |
GPT-5 Chat Chat-optimized variant of GPT-5 for conversational use. | Chat | $1.25 | $10.00 | 128K | |
GPT-5 Codex Code-generation focused variant of GPT-5. | Coding | $1.50 | $12.00 | 256K | |
Computer Use Preview Model specialized in GUI interaction and computer control. | Agentic | $3.00 | $12.00 | 128K | |
GPT-4.5 Preview Earlier preview model. Superseded by GPT-4.1 and GPT-5. | Legacy | $30.00 | $60.00 | 128K | |
DALL·E 3 Image generation model. Priced per image, not per token. | Image | $0.040 | $0.040 | — | |
| Anthropic | Claude Opus 4.8 Current Anthropic flagship for coding, long-context, and multi-step agent loops. 1M context at significantly lower cost than prior Opus tiers. | Flagship | $5.00 | $25.00 | 1.0M |
Claude Opus 5.5 First model in Anthropic's Claude 5.5 family (Sep 22, 2026). Fable 5.1-level performance at ~40% lower workload cost than Opus 5, with a 60% cache-read price cut. Targets long-running agentic coding, computer use, and knowledge work. BenchAlign v5.6 score 88.5. | Flagship | $4.00 | $20.00 | 1.0M | |
Claude 4 Opus Most capable Claude model for complex agentic workflows. | Flagship | $15.00 | $75.00 | 200K | |
Claude 4 Sonnet Latest Sonnet with extended thinking. Best price-performance. | Mid-range | $3.00 | $15.00 | 200K | |
Claude 4 Haiku Fast, cost-effective model for high-volume tasks. | Budget | $0.625 | $2.50 | 200K | |
Claude 3.7 Sonnet Prior-generation Sonnet with extended thinking mode. | Mid-range | $3.00 | $15.00 | 200K | |
Claude 3.5 Sonnet Previous-generation Sonnet. Still strong for coding tasks. | Legacy | $3.00 | $15.00 | 200K | |
Claude 3.5 Haiku Previous-generation Haiku. Good for fast, simple tasks. | Legacy | $0.800 | $4.00 | 200K | |
Claude 3 Opus Original Opus model. Superseded by Claude 4 Opus. | Legacy | $15.00 | $75.00 | 200K | |
Gemini 4 Argon Public id gemini-4-argon. Intro rate $2/$10 per million tokens through November 30, 2026, then $4/$20, from the Oct 13 blog footnote and the first Agent Platform table. Cached input on that table is $0.10, then $0.20. Two other Argon tables on the same page had no heading in the fetch. Explicit caching, image generation, and Live API are unsupported on the model docs page. Default sync output cap is 65,536; long decode can use the unified 1,048,576. | Flagship | $2.00 | $10.00 | 1.0M | |
EmbeddingGemma 2 740M open multimodal embedding model. Maps text, images, video, and audio into one 768-d space. No token price on the October 6, 2026 card. | Embedding | — | — | 8K | |
Gemini 3.1 Pro Google's high-capability multimodal model for long context, coding agents, and Vertex / AI Studio deployments. | Flagship | $2.50 | $15.00 | 1.0M | |
Gemini 2.5 Pro 1M token context. Strong coding and reasoning. Best Gemini model. | Flagship | $1.25 | $10.00 | 1.0M | |
Gemini 2.5 Flash Fast, cost-efficient Gemini with 1M context window. | Mid-range | $0.150 | $0.600 | 1.0M | |
Gemini 2.0 Flash Stable 2.0 Flash for production workloads. | Mid-range | $0.100 | $0.400 | 1.0M | |
Gemini 2.0 Flash Lite Ultra-lightweight model for high-volume tasks. | Budget | $0.075 | $0.300 | 1.0M | |
Gemini 1.5 Pro 2M context window model. Good for long document analysis. | Legacy | $1.25 | $5.00 | 2.1M | |
Gemini 1.5 Flash Fast, affordable 1M context model. Superseded by 2.5 Flash. | Legacy | $0.075 | $0.300 | 1.0M | |
Gemma 3 27B Open-weight instruction-tuned model. Free to download and run locally. | Open Source | $0.000 | $0.000 | 128K | |
Gemma 3 12B Mid-size open-weight model. Good balance for local deployment. | Open Source | $0.000 | $0.000 | 128K | |
Gemma 3 4B Small open-weight model. Runs on consumer hardware. | Open Source | $0.000 | $0.000 | 128K | |
Gemma 4 QAT 32B Quantization-aware training release. 72% less VRAM needed. Runs on 16GB GPUs. | Open Source | $0.000 | $0.000 | 256K | |
Gemma 4 QAT 14B Mid-size QAT model. Efficient inference for 8GB GPUs. | Open Source | $0.000 | $0.000 | 128K | |
Imagen 3 Text-to-image generation model. Priced per image. | Image | $0.020 | $0.020 | — | |
| DeepSeek | DeepSeek-V4.1-Flash DeepSeek's Causal Encoder-Decoder MoE (552B total, 8B active prefill) cuts KV cache 8x. GPQA Diamond 90.9, Terminal-Bench 2.1: 90.6. $0.003/M input off-peak. MIT-licensed open weights. | Flagship | $0.003 | $0.210 | 400K |
DeepSeek-V4 Latest DeepSeek flagship. 1M context, top-tier coding and math. | Flagship | $0.300 | $0.600 | 1.0M | |
DeepSeek-V4 Flash Fast variant of V4. Incredible price-performance ratio. | Mid-range | $0.140 | $0.280 | 1.0M | |
DeepSeek-V4 Pro DeepSeek V4 Pro reasoning model. Replaces deepseek-reasoner. | Reasoning | $0.435 | $0.870 | 1.0M | |
DeepSeek-V3 Strong open-weight model at very low cost. | Legacy | $0.140 | $0.280 | 64K | |
DeepSeek-R1 Reasoning model. Output includes chain-of-thought. | Reasoning | $0.550 | $2.19 | 64K | |
DeepSeek-R1-0528 Updated R1 with improved coding and longer output. | Reasoning | $0.550 | $2.19 | 64K | |
| xAI | Grok 3 xAI flagship with real-time X data access. | Flagship | $3.00 | $15.00 | 131K |
Grok 4.7 xAI's September 21, 2026 flagship for coding and knowledge work. Larger base model with longer RL training on multi-hour tasks. Held at Grok 4.6's $2/$6 token rates, but independent testing shows ~2.5x higher token use. Artificial Analysis Intelligence Index 46 (xHigh). 500K context, multimodal input. | Flagship | $2.00 | $6.00 | 500K | |
Grok 3 Mini Fast, affordable Grok model for everyday tasks. | Budget | $0.300 | $0.500 | 131K | |
Grok 2 Previous-generation Grok. Still capable for general tasks. | Legacy | $2.00 | $10.00 | 131K | |
Grok 2 Mini Budget previous-generation Grok model. | Legacy | $0.200 | $0.600 | 131K | |
| Meta | Muse Glimmer 30B Meta's 30B open-weight agentic model (Apache 2.0). Runs on a single consumer GPU with DFlash speculative decoding. 131K context, multimodal. | Open Source | $0.000 | $0.000 | 131K |
Muse Spark 1.2 Meta's frontier coding model co-trained with Muse Code agent. Improved agentic knowledge-work performance. No published benchmark scores at launch. | Flagship | $3.00 | $6.00 | 128K | |
Llama 4 Maverick Open-weight multimodal MoE model. 128 active of 400B params. | Flagship | $0.200 | $0.600 | 256K | |
Llama 4 Scout Efficient MoE model. 1M context window. Good for RAG. | Mid-range | $0.150 | $0.450 | 1.0M | |
Llama 3.3 70B Stable 70B model. Good balance of capability and speed. | Mid-range | $0.880 | $0.880 | 131K | |
Llama 3.1 405B Largest open-weight model. Requires significant compute. | Flagship | $3.00 | $3.00 | 131K | |
Llama 3.1 70B Popular 70B model for fine-tuning and local deployment. | Mid-range | $0.880 | $0.880 | 131K | |
Llama 3.1 8B Small, fast model. Great for edge devices and prototyping. | Budget | $0.050 | $0.050 | 131K | |
Llama 3.2 11B Vision Multimodal model with image understanding. | Multimodal | $0.180 | $0.180 | 131K | |
Llama 3.2 90B Vision Large multimodal model with strong vision capabilities. | Multimodal | $0.900 | $0.900 | 131K | |
| Mistral | Mistral Large 4 Public preview. 1 trillion parameters, 49 billion active. Card rate $1.36 / $4.18 per million tokens. Weights not released on the October 6, 2026 post. Context window not published. | Flagship | $1.36 | $4.18 | — |
Mistral Large Mistral's most capable model. Strong multilingual and coding. | Flagship | $2.00 | $6.00 | 128K | |
Mistral Medium Good balance of cost and capability for production use. | Mid-range | $0.400 | $2.00 | 128K | |
Mistral Small Fast, affordable model for high-volume tasks. | Budget | $0.200 | $0.600 | 128K | |
Mistral Nemo Open-weight 12B model. Good for fine-tuning. | Open Source | $0.150 | $0.150 | 128K | |
Mistral Codestral Code generation model. 256K context, 80+ languages. | Coding | $0.300 | $0.900 | 256K | |
Pixtral Large Multimodal model combining vision and text understanding. | Multimodal | $2.00 | $6.00 | 128K | |
Mistral Embed Embedding model for search and RAG applications. | Embedding | $0.100 | $0.000 | 8K | |
| Cohere | Command R+ Cohere's most capable model. Strong for RAG and enterprise search. | Flagship | $2.50 | $10.00 | 128K |
Command R Cost-effective model optimized for RAG and tool use. | Mid-range | $0.500 | $2.00 | 128K | |
Command A Latest Command model with extended context and improved performance. | Mid-range | $0.950 | $3.80 | 256K | |
Command R7B Small, fast model for classification and simple RAG. | Budget | $0.050 | $0.200 | 128K | |
Cohere Embed v3 Multilingual embedding model for search and retrieval. | Embedding | $0.100 | $0.000 | 512 | |
Cohere Rerank v3 Reranking model for improving search relevance. | Reranking | $0.100 | $0.000 | 4K | |
| Moonshot AI | Kimi K2 Latest Kimi model. Strong coding and agentic capabilities. MoE architecture. | Flagship | $0.600 | $2.40 | 131K |
Kimi K2 Flash Fast Kimi variant. Great price-performance for everyday tasks. | Mid-range | $0.150 | $0.600 | 131K | |
Moonshot v1 128K Previous-generation Kimi with 128K context. | Legacy | $0.840 | $3.36 | 131K | |
Moonshot v1 32K Budget Kimi variant with 32K context. | Legacy | $0.480 | $1.92 | 33K | |
Moonshot v1 8K Lowest-cost Kimi variant for short-context tasks. | Budget | $0.240 | $0.960 | 8K | |
| MiniMax | MiniMax-M2.7 Latest MiniMax model. 1M context, strong multilingual. | Flagship | $0.300 | $1.20 | 1.0M |
MiniMax-Text-01 Budget MiniMax model. Good for high-volume tasks. | Mid-range | $0.100 | $0.400 | 1.0M | |
Hailuo 2.3 (Video) Video generation model. Pricing per second of video. | Video | $0.050 | $0.050 | — | |
Speech 2.8 Text-to-speech model with voice cloning. | Speech | $0.010 | $0.000 | — | |
Music 2.5 Music generation from text prompts. | Music | $0.050 | $0.000 | — | |
| Alibaba Cloud (Qwen) | Qwen3 235B-A22B MoE model. 235B total, 22B active. Top open-source coding model. | Flagship | $0.800 | $1.60 | 128K |
Qwen3 32B Dense 32B model. Great for local deployment. | Mid-range | $0.300 | $0.600 | 128K | |
Qwen3 14B Compact model. Runs on consumer GPUs. | Budget | $0.150 | $0.300 | 128K | |
Qwen3 4B Small model for edge devices and mobile. | Budget | $0.050 | $0.100 | 33K | |
Qwen3 1.7B Ultra-small model. Good for classification and extraction. | Budget | $0.010 | $0.020 | 33K | |
Qwen3 Coder 480B-A35B MoE coding specialist. 480B total, 35B active params. | Coding | $0.500 | $1.00 | 256K | |
Qwen2.5 72B Previous-generation 72B. Still strong for many tasks. | Legacy | $0.400 | $0.800 | 131K | |
Qwen2.5-VL 72B Vision-language model. Image and video understanding. | Multimodal | $0.500 | $1.00 | 131K | |
Qwen3.6-27B Dense 27B model for coding, reasoning, and agent workloads. | Mid-range | $0.200 | $0.400 | 131K | |
Qwen-AgentWorld-35B-A3B Qwen's language world model that simulates agent environments across MCP, search, terminal, SWE, Android, web, and OS in a single 35B MoE (3B active). Apache 2.0. | Agentic | $0.300 | $1.20 | 262K | |
| NVIDIA | Nemotron 3 Ultra 253B NVIDIA's flagship MoE model. 253B params, 55B active. | Flagship | $0.800 | $2.40 | 131K |
Llama 3.3 70B (NIM) Meta's Llama 3.3 70B served via NVIDIA NIM. | Hosted | $0.700 | $0.700 | 131K | |
DeepSeek R1 (NIM) DeepSeek R1 reasoning model served via NIM. | Hosted | $0.880 | $3.52 | 131K | |
Qwen 2.5 72B (NIM) Qwen 2.5 72B served via NVIDIA NIM. | Hosted | $0.700 | $0.700 | 131K | |
Mistral Small (NIM) Mistral Small served via NVIDIA NIM. | Hosted | $0.200 | $0.200 | 131K | |
| Ollama | Gemma 4 QAT 32B Google's latest. 72% less VRAM. Runs on 16GB GPUs. | Featured | $0.000 | $0.000 | 256K |
Gemma 4 QAT 14B Mid-size Gemma 4 QAT. Efficient for 8GB GPUs. | Featured | $0.000 | $0.000 | 128K | |
Qwen3 235B-A22B MoE model. 22B active params. Strong coding and reasoning. | Featured | $0.000 | $0.000 | 131K | |
Qwen3 32B Best open-source 32B model. Great for local deployment. | Popular | $0.000 | $0.000 | 131K | |
DeepSeek R1 Open-weight reasoning model. Chain-of-thought included. | Reasoning | $0.000 | $0.000 | 131K | |
DeepSeek V3 671B MoE model. Top open-source general model. | Popular | $0.000 | $0.000 | 66K | |
Llama 4 Maverick Meta's latest MoE. 128 active of 400B params. | Popular | $0.000 | $0.000 | 256K | |
Llama 3.3 70B Stable workhorse 70B model for production. | Popular | $0.000 | $0.000 | 131K | |
Gemma 3 27B Google's open-weight 27B. Strong for its size. | Popular | $0.000 | $0.000 | 128K | |
Mistral Small 3.1 24B Mistral's latest small model. Vision support included. | Popular | $0.000 | $0.000 | 131K | |
Phi-4 14B Microsoft's compact reasoning model. Punches above its weight. | Popular | $0.000 | $0.000 | 16K | |
Command R Cohere's open-weight RAG model. Built for retrieval. | RAG | $0.000 | $0.000 | 131K | |
Codestral 22B Mistral's open-weight coding model. 80+ languages. | Coding | $0.000 | $0.000 | 66K | |
GLM-4 9B Zhipu AI's open-weight bilingual model. | Popular | $0.000 | $0.000 | 131K | |
Llama 3.1 8B Smallest Llama 3.1. Great for fine-tuning. | Budget | $0.000 | $0.000 | 131K | |
Gemma 3 4B Compact Gemma for edge devices. | Budget | $0.000 | $0.000 | 131K | |
Qwen3 4B Small Qwen for mobile and edge. | Budget | $0.000 | $0.000 | 33K | |
SmolLM2 1.7B HuggingFace's tiny model. Runs on phones. | Edge | $0.000 | $0.000 | 8K | |
Nomic Embed Text Open-weight embedding model for search and RAG. | Embedding | $0.000 | $0.000 | 8K | |
Stable Diffusion 3.5 Open-weight image generation model. | Image | $0.000 | $0.000 | — | |
| OpenRouter | Claude 4 Opus Best Claude model. Full power for complex tasks. | Flagship | $15.00 | $75.00 | 200K |
Claude 4 Sonnet Best price-performance Claude. | Popular | $3.00 | $15.00 | 200K | |
GPT-5.2 Latest GPT model via OpenRouter. | Flagship | $2.50 | $15.00 | 256K | |
Gemini 2.5 Pro Google's best model with 1M context. | Popular | $1.25 | $10.00 | 1.0M | |
Gemini 2.5 Flash Fast, affordable Gemini. | Popular | $0.150 | $0.600 | 1.0M | |
DeepSeek V4 Incredible price-performance. 1M context. | Popular | $0.300 | $0.600 | 1.0M | |
o3 OpenAI's reasoning model for hard problems. | Reasoning | $10.00 | $40.00 | 200K | |
o4-mini Affordable reasoning model. | Reasoning | $1.10 | $4.40 | 200K | |
Llama 4 Maverick Meta's latest open-weight model. | Open Source | $0.200 | $0.600 | 256K | |
Kimi K2 Moonshot's latest coding model. MoE architecture. | Popular | $0.600 | $2.40 | 131K | |
Qwen3 235B-A22B Top open-source MoE model. | Open Source | $0.800 | $1.60 | 128K | |
Mistral Large Mistral's most capable model. | Popular | $2.00 | $6.00 | 128K | |
Grok 3 xAI's flagship with real-time data access. | Popular | $3.00 | $15.00 | 131K | |
Command A Cohere's latest model for RAG and enterprise. | RAG | $0.950 | $3.80 | 256K | |
MiniMax-M2.7 MiniMax with 1M context. | Popular | $0.300 | $1.20 | 1.0M | |
GPT-4.1 Long-context coding model. | Coding | $2.00 | $8.00 | 1.0M | |
GPT-4.1 Mini Affordable long-context model. | Popular | $0.400 | $1.60 | 1.0M | |
GPT-5 Nano Ultra-cheap for classification and routing. | Budget | $0.050 | $0.300 | 128K | |
DeepSeek R1 Open-weight reasoning model. | Reasoning | $0.550 | $2.19 | 64K | |
Nemotron 3 Ultra 253B NVIDIA's flagship MoE model. | New | $0.800 | $2.40 | 131K | |
| Together AI | Llama 4 Maverick Meta's latest MoE on fast serverless inference. | Flagship | $0.180 | $0.540 | 1.0M |
Llama 3.3 70B Stable 70B on fast inference. | Popular | $0.880 | $0.880 | 131K | |
Qwen3 235B-A22B Qwen's MoE model on Together. | Popular | $0.800 | $1.60 | 131K | |
Qwen3 32B Dense 32B at low cost. | Popular | $0.300 | $0.600 | 131K | |
DeepSeek R1 DeepSeek reasoning on Together. | Reasoning | $0.550 | $2.19 | 131K | |
DeepSeek V3 672B MoE at unbeatable pricing. | Popular | $0.140 | $0.280 | 66K | |
Gemma 3 27B Google's 27B open-weight on Together. | Popular | $0.240 | $0.240 | 131K | |
Mistral Small 3.1 24B Mistral's latest small model. | Popular | $0.100 | $0.100 | 131K | |
Codestral 22B Mistral's coding specialist. | Coding | $0.150 | $0.450 | 66K | |
FLUX.2 Pro Top open-weight image model. $0.03/MP. | Image | $0.030 | $0.030 | — | |
FLUX.2 Dev Open-weight image generation. Great quality per dollar. | Image | $0.010 | $0.010 | — | |
| Shanghai AI Laboratory | Atria Dawn Preview 744B MoE agentic model with MIT weights and 1M context. Top scores on BrowseComp (92.5), CyberGym (86.5), AutomationBench (53.8). No API — self-host only. Trained via Verifiable Experience Pipeline. | Open Source | — | — | 1.0M |
| Aleph Alpha | Kolibri 1 78B total, 3.46B active. Card validates 1,048,576 tokens and recommends serving at or under 262,144. No public token price. Apache 2.0 weights only. | Open Source | — | — | 1.0M |
| Cloudflare | Clef 27B decision model at $0.24 per million input tokens. Scores a typed schema. Does not generate free-form text. No output price published. | Decision | $0.240 | — | 66K |
Clef-flash 9B decision model at $0.09 per million input tokens. Same request schema as Clef. No output price published. | Decision | $0.090 | — | 66K | |
Clef-Omni 30B-A3B decision model. Workers AI model page lists $0.15 per million input tokens and no output charge. The pricing table fetched 2026-10-14 did not list this row. Scores a typed schema over text, images, audio, or video. Does not generate free-form text. | Decision | $0.150 | — | 64K | |
| Perplexity | pplx-decider-v1-27b 27B decision model. Hub record created 2026-10-01 and duplicated from denis-pplx/autojev-27b. No token price or context length on the card. | Decision | — | — | — |
Source: provider pricing pages. Prices are per 1M tokens unless noted. Benchmarks from public leaderboards. Free models ($0) are open-weight — run locally or via hosted platform.