The Open-Source AI Stack
RSS

Models

What's actually shipped, and how it compares

One row per checkpoint, not per family. 41 OSI-open, 34 source-available open-weights, 40 proprietary baselines, since Feb 2023. Every architecture parameter, benchmark score, release date, and price is sourced.

How to read model names expand ↓

Model names look like jargon because labs cram several facts into one string. Once you know the convention they're easier to read than they look.

  • Llama 3.3 70B Instruct

    Llama = family · 3.3 = version · 70B = parameter count (70 billion) · Instruct = post-trained for following instructions (vs "Base" which is the pretrained checkpoint before any RLHF).

  • DeepSeek-R1

    DeepSeek = family · R1 = "Reasoning 1", a model post-trained with reinforcement learning to emit chain-of-thought before answering. The "thinking model" naming convention started with OpenAI's o1; R1 is DeepSeek's open answer to it.

  • Qwen 3 235B A22B Instruct

    Qwen = family · 3 = version · 235B = total parameters across all experts · A22B = "active 22 billion" — the parameters used per forward pass. This is the MoE convention: total/active. Compare to DeepSeek V3 at 671B total / 37B active.

  • Mixtral 8x7B Instruct v0.1

    Mixtral = family · 8x7B = eight 7B experts, with two activated per token (Mistral's earlier MoE notation; same idea as Qwen's total/active just expressed differently) · v0.1 = release version.

  • Claude 3.5 Sonnet

    Claude = family · 3.5 = version · Sonnet = tier name (Anthropic uses Haiku < Sonnet < Opus to rank speed/cost/quality within a generation; tier semantics shift across generations).

  • GPT-4o vs o1 vs o3

    GPT-4o is OpenAI's multimodal model. The o-series (o1, o3, o3-mini) is their reasoning line — the o stands for "OpenAI" branding here, not a version qualifier. Numbers grow with capability, not with chronology (o3 launched before o2 was named publicly).

Launch timeline

open open-weights source-available proprietary 2023 2024 2025 2026

Dots in lanes by openness, colored by family. Hover for details; click to open the model page.

claude (14) llama (13) deepseek (10) mistral (6) qwen (6) gemini (4) gemma (4) gpt-4 (4) grok (4) o-series (4) command (3) granite (3) olmo (3) phi (3) falcon (1) glm (1) gpt-3-5 (1) kimi (1) starcoder (1) tulu (1) yi (1) other (27)

Compare any two metrics

Pick an X and Y metric; one dot per model. Cost and parameter axes use log scale automatically. Open as filled dot, closed as ring. Models that did not publish the picked metric drop off the chart. Costs and speeds via Artificial Analysis ↗; benchmark scores from each model's primary source.

Drag to pan. Scroll to zoom. Shift + drag to box-zoom. Click a dot to open the model page.

Why some models are missing from the plot

All models (115)

115 models
cmp Model Developer Released ▼ Openness Architecture Total params Active params Context $/M in $/M out tok/s MMLU GPQA-D HumanEval Quants
Command A+ Cohere 2026-05-20 Open moe 218B 25B 128K $0.00 $0.00 212.2 — 76.1 — gguf fp8
Grok 4.3 xAI — Proprietary unknown — — — $1.25 $2.50 88.1 — 90.1 — —
Granite 4.1 8B Instruct IBM 2026-04-29 Open dense 8B — 524K — — — — — — gguf awq mlx fp8
DeepSeek-V4 Pro DeepSeek — Open moe 1.6T 49B 1.0M $1.74 $3.48 29.8 — 88.8 76.8 gguf mlx fp8
GPT-5.5 OpenAI 2026-04-23 Proprietary unknown — — 1.0M $5.00 $30.00 66.7 — 93.5 — —
Kimi K2.6 Moonshot AI 2026-04-21 Open moe 1T 32B 262K $0.95 $4.00 63.9 — 90.5 — gguf mlx fp8
Claude Opus 4.7 Anthropic 2026-04-16 Proprietary unknown — — 1.0M $5.00 $25.00 48.6 — 91.4 — —
Gemma 4 31B Google DeepMind 2026-04-02 Open weights dense 31B — 262K $0.00 $0.00 35.5 — 85.7 — gguf awq gptq exl2 mlx fp8 bnb
Gemma 4 26B-A4B Google DeepMind 2026-04-02 Open weights moe 26B 3.8B 262K — — — — — — gguf awq gptq exl2 mlx fp8
Gemini 3.1 Pro Google DeepMind 2026-02-19 Proprietary unknown — — 1.0M — — — — — — —
Claude Sonnet 4.6 Anthropic 2026-02-17 Proprietary unknown — — 1.0M $3.00 $15.00 49 — 79.9 — —
Qwen3.5 Plus Alibaba 2026-02-16 Open moe 397B 17B 1.0M — — — — — — —
Qwen3.5 122B-A10B Alibaba 2026-02-16 Open moe 122B 10B 262K — — — — — — gguf awq gptq mlx fp8
Qwen3.5 35B-A3B Alibaba 2026-02-16 Open moe 35B 3B 262K — — — — — — gguf awq gptq mlx fp8
Claude Opus 4.6 Anthropic 2026-02-05 Proprietary unknown — — 1.0M $5.00 $25.00 43.7 — 84.0 — —
Kimi K2.5 Moonshot AI 2026-01-27 Open moe 1T 32B 256K $0.58 $3.00 33.9 — 87.6 — gguf mlx fp8
Gemini 3 Flash Google DeepMind 2025-12-17 Proprietary unknown — — 1.0M $0.50 $3.00 185.9 — 90.4 — —
Nemotron 3 Nano NVIDIA 2025-12-15 Open weights hybrid-mamba-transformer-moe 30B — 1.0M — — — — — — gguf awq gptq mlx fp8
Tencent Hunyuan 2.0 Instruct Tencent 2025-12-05 Open weights moe 406B 32B 256K — — — — — — —
Mistral Large 3 Mistral AI 2025-12-02 Open weights moe 675B 41B 256K $0.50 $1.50 54.2 — 68.0 — gguf
DeepSeek-V3.2 DeepSeek — Open moe — — — $0.50 $1.60 0 — 75.1 — gguf awq mlx fp8
Claude Opus 4.5 Anthropic 2025-11-24 Proprietary unknown — — — $5.00 $25.00 51.8 — 81.0 — —
OLMo 3 32B Instruct AI2 2025-11-20 Open dense 32B — 66K — — — 85.4 — — gguf awq gptq mlx fp8
Gemini 3 Pro Google DeepMind 2025-11-18 Proprietary unknown — — 1.0M $2.00 $12.00 132.9 — 91.9 — —
Grok 4.1 xAI — Proprietary unknown — — — — — — — — — —
GPT-5.1 OpenAI 2025-11-12 Proprietary unknown — — 400K $1.25 $10.00 114.7 — 87.3 — —
Kimi K2 Thinking Moonshot AI 2025-11-06 Open moe 1T 32B 256K $0.60 $2.50 102.4 — 84.5 — gguf mlx fp8 bnb
Claude Haiku 4.5 Anthropic 2025-10-15 Proprietary unknown — — 200K $1.00 $5.00 — — — — —
Granite 4.0 Small IBM 2025-10-02 Open hybrid-mamba-transformer-moe 32B 9B 131K — — — — — — gguf awq mlx fp8 bnb
GLM-4.6 Zhipu AI 2025-09-30 Open moe 357B — 128K $0.60 $2.20 30.7 — 63.2 — gguf awq gptq exl2 mlx fp8
Claude Sonnet 4.5 Anthropic 2025-09-29 Proprietary unknown — — — $3.00 $15.00 48.8 — 72.7 — —
Qwen3 Max Alibaba — Proprietary moe — — — $1.66 $7.22 32.4 — 76.4 — —
Qwen3-VL 235B-A22B Instruct Alibaba 2025-09-23 Open moe 235B 22B 262K $0.30 $1.90 50.9 — 71.2 — gguf awq mlx fp8
Qwen3 Next 80B-A3B Instruct Alibaba — Open moe 80B 3B 262K — — — — — — gguf awq gptq exl2 mlx fp8 bnb
DeepSeek-V3.1 DeepSeek — Open moe 671B 37B 128K $0.56 $1.67 0 — 73.5 — gguf awq gptq mlx fp8 bnb
GPT-5 OpenAI — Proprietary unknown — — — $1.25 $10.00 72 — 85.4 — —
Claude Opus 4.1 Anthropic 2025-08-05 Proprietary unknown — — — — — — — — — —
GLM-4.5 Zhipu AI 2025-07-28 Open moe 355B 32B 128K — — — — — — gguf awq gptq exl2 mlx
Kimi K2 Instruct Moonshot AI 2025-07-11 Open moe 1T 32B 128K — — — — 75.1 — gguf gptq mlx fp8 bnb
Grok 4 xAI — Proprietary unknown — — — $5.50 $27.50 0 — 87.7 — —
Apple On-Device Foundation Model (2025) Apple 2025-06-09 Proprietary dense 3B — — — — — — — — —
DeepSeek-R1 (May 2025 refresh) DeepSeek 2025-05-28 Open moe — — — — — — — 81.0 — gguf awq gptq mlx fp8 bnb
Claude Sonnet 4 Anthropic 2025-05-22 Proprietary unknown — — — $3.00 $15.00 48.4 — 70.0 — —
Claude Opus 4 Anthropic 2025-05-22 Proprietary unknown — — — $15.00 $75.00 38.1 — 74.9 — —
Mistral Medium 3 Mistral AI 2025-05-07 Proprietary unknown — — 128K $0.40 $2.00 29 — 57.8 — —
Phi-4 Reasoning Microsoft 2025-04-30 Open dense 14B — 32K — — — — 65.8 — gguf awq mlx fp8 bnb
Qwen 3 235B A22B Instruct Alibaba 2025-04-28 Open moe 235B 22B 131K $0.45 $1.80 66.6 — — — gguf awq gptq exl2 mlx fp8
Qwen 3 32B Instruct Alibaba 2025-04-28 Open dense 32.8B — 131K $0.15 $0.59 98.7 — 53.5 — gguf awq gptq exl2 mlx fp8 bnb
Gemini 2.5 Flash Google DeepMind — Proprietary unknown — — 1.0M $0.30 $2.50 196.8 — 68.3 — —
OpenAI o3 OpenAI 2025-04-16 Proprietary unknown — — 200K $2.00 $8.00 88 — 87.7 — —
OpenAI o4-mini OpenAI 2025-04-16 Proprietary unknown — — 200K $1.10 $4.40 151.2 — 78.4 — —
Llama-3.1-Nemotron Ultra 253B v1 NVIDIA 2025-04-11 Open weights dense 253B — 131K — — — — 76.0 — gguf mlx fp8
Llama 4 Scout Meta 2025-04-05 Source-available moe 109B 17B 10.5M $0.17 $0.66 108.1 — — — gguf awq mlx fp8 bnb
Llama 4 Maverick Meta 2025-04-05 Source-available moe 400B 17B — $0.35 $0.85 108.9 — — — gguf awq mlx fp8
Gemini 2.5 Pro Google DeepMind 2025-03-25 Proprietary unknown — — 1.0M $1.25 $10.00 127.3 — — — —
Llama-3.3-Nemotron Super 49B v1 NVIDIA 2025-03-18 Open weights dense 49B — 131K $0.00 $0.00 0 — 66.7 — gguf mlx fp8
Command A Cohere 2025-03-13 Open weights dense 111B — 256K $2.50 $10.00 44.8 — 52.7 — gguf awq gptq exl2 mlx fp8 bnb
Gemma 3 27B IT Google DeepMind 2025-03-12 Source-available dense 27B — 131K — — — — — 87.8 gguf awq gptq exl2 mlx fp8 bnb
GPT-4.5 OpenAI 2025-02-27 Proprietary unknown — — 128K $0.00 $0.00 0 — — — —
Phi-4-mini Instruct Microsoft — Open dense — — 128K — — — 67.3 — — gguf awq gptq mlx fp8 bnb
Claude 3.7 Sonnet Anthropic 2025-02-24 Proprietary unknown — — — $3.00 $15.00 0 — — — —
Grok 3 xAI 2025-02-17 Proprietary unknown — — 131K $4.00 $20.00 0 — 69.3 — —
OpenAI o3-mini OpenAI 2025-01-31 Proprietary unknown — — 200K $1.10 $4.40 145.3 — 79.7 — —
DeepSeek-R1 DeepSeek — Open moe 671B 37B 128K $1.35 $4.20 0 90.8 71.5 — gguf awq gptq mlx fp8
DeepSeek-R1-Distill-Llama-70B DeepSeek 2025-01-20 Open dense 70B — 128K $0.70 $1.05 43.5 — 65.2 — gguf awq gptq exl2 mlx fp8 bnb
DeepSeek-R1-Distill-Qwen-32B DeepSeek 2025-01-20 Open dense 32B — 128K $0.00 $0.00 0 — 62.1 — gguf awq gptq exl2 mlx fp8 bnb
DeepSeek-V3 DeepSeek — Open moe 671B 37B 128K $0.40 $0.89 0 87.1 59.1 65.2 gguf awq gptq mlx fp8
Phi-4 Microsoft 2024-12-12 Open dense 14B — 16K $0.13 $0.50 30.9 84.8 56.1 82.6 gguf awq gptq exl2 mlx fp8 bnb
Gemini 2.0 Flash Google DeepMind 2024-12-11 Proprietary unknown — — 1.0M $0.15 $0.60 0 — 62.3 — —
Llama 3.3 70B Instruct Meta 2024-12-06 Source-available dense 70B — 131K — — — 86.0 50.5 88.4 gguf awq gptq exl2 mlx fp8 bnb
OpenAI o1 OpenAI — Proprietary unknown — — — $15.00 $60.00 75.8 — 77.3 — —
QwQ 32B Preview Alibaba 2024-11-27 Open dense 32.5B — 33K — — — — — — gguf awq gptq exl2 mlx fp8 bnb
OLMo 2 7B Instruct AI2 2024-11-26 Open dense 7B — — — — — 61.3 — — gguf awq gptq mlx
OLMo 2 13B Instruct AI2 2024-11-26 Open dense 13.7B — 4K — — — 68.5 — — gguf awq gptq mlx
Llama 3.1 Tülu 3 70B AI2 2024-11-21 Open weights dense 70B — — — — — 83.1 — 92.4 gguf awq exl2 mlx
Qwen 2.5 Coder 32B Instruct Alibaba 2024-11-12 Open dense 32.5B — 131K $0.00 $0.00 0 — 41.7 — gguf awq gptq exl2 mlx fp8 bnb
Claude 3.5 Haiku Anthropic 2024-11-04 Proprietary unknown — — 200K $0.80 $4.00 0 — 40.8 — —
Granite 3.0 8B Instruct IBM 2024-10-21 Open dense 8.1B — 4K — — — 65.8 33.8 64.6 gguf awq
Llama 3.2 1B Instruct Meta 2024-09-25 Source-available dense 1.2B — 131K — — — — — — gguf awq gptq exl2 mlx fp8 bnb
Llama 3.2 3B Instruct Meta 2024-09-25 Source-available dense 3.2B — 131K — — — — — — gguf awq gptq exl2 mlx fp8 bnb
Llama 3.2 11B Vision Instruct Meta 2024-09-25 Source-available dense — — 131K — — — — — — gguf gptq fp8 bnb
Llama 3.2 90B Vision Instruct Meta 2024-09-25 Source-available dense — — 131K — — — — — — fp8 bnb
Qwen 2.5 72B Instruct Alibaba 2024-09-19 Open dense 72.7B — 131K $0.36 $0.40 54.1 — 49.0 86.6 gguf awq gptq exl2 mlx fp8 bnb
DeepSeek-V2.5 DeepSeek — Open weights moe — — — $0.00 $0.00 0 — — 89.0 gguf mlx
Mistral Large 2 Mistral AI 2024-07-24 Source-available dense 123B — 131K $2.00 $6.00 31.7 84.0 48.6 92.0 gguf gptq exl2
Llama 3.1 8B Instruct Meta 2024-07-23 Source-available dense 8.0B — 131K — — — — 30.4 — gguf awq gptq exl2 mlx fp8 bnb
Llama 3.1 70B Instruct Meta 2024-07-23 Source-available dense 70B — 131K — — — — — — gguf awq gptq exl2 mlx fp8 bnb
Llama 3.1 405B Instruct Meta 2024-07-23 Source-available dense 405B — 131K — — — — — — gguf awq gptq mlx fp8 bnb
Mistral Nemo 12B Instruct Mistral AI 2024-07-18 Open dense 12B — 131K — — — 68.0 — — gguf awq exl2 mlx fp8 bnb
Gemma 2 27B IT Google DeepMind 2024-06-27 Source-available dense 27.2B — 8K — — — — — — gguf awq mlx fp8 bnb
Claude 3.5 Sonnet Anthropic 2024-06-20 Proprietary unknown — — 200K — — — — — — —
DeepSeek-Coder-V2 Instruct DeepSeek 2024-06-17 Open weights moe 236B 21B 128K — — — — — 90.2 gguf mlx
Qwen 2 72B Instruct Alibaba 2024-06-06 Source-available dense 72.7B — 131K $0.00 $0.00 0 82.3 37.1 86.0 gguf awq gptq exl2 mlx fp8
GLM-4-9B-Chat Zhipu AI 2024-06-05 Source-available dense 9B — 131K — — — 72.4 — 71.8 gguf
Codestral 22B v0.1 Mistral AI 2024-05-29 Source-available dense 22B — 33K — — — — — — gguf exl2 mlx
Phi-3 Medium 4K Instruct Microsoft 2024-05-21 Open dense 14B — 4K — — — 78.0 — 62.2 gguf awq
GPT-4o OpenAI 2024-05-13 Proprietary unknown — — — $2.50 $10.00 131.6 — 54.3 — —
DeepSeek-V2 Chat DeepSeek 2024-05-07 Open moe 236B 21B 128K — — — — — — gguf mlx
Phi-3 Mini 4K Instruct Microsoft 2024-04-23 Open dense 3.8B — 4K — — — 70.9 30.6 57.3 gguf awq mlx bnb
Llama 3 8B Instruct Meta 2024-04-18 Source-available dense 8B — 8K — — — — — — gguf awq gptq exl2 mlx fp8 bnb
Llama 3 70B Instruct Meta 2024-04-18 Source-available dense — — 8K — — — — — — gguf awq gptq exl2 fp8 bnb
Mixtral 8x22B Instruct v0.1 Mistral AI 2024-04-17 Open moe 141B 39B 66K — — — — — — gguf awq mlx fp8
GPT-4 Turbo OpenAI 2024-04-09 Proprietary unknown — — 128K $10.00 $30.00 27.8 — — — —
Command R+ Cohere 2024-04-04 Source-available dense — — — — — — — — — gguf fp8
Claude 3 Haiku Anthropic 2024-03-13 Proprietary unknown — — 200K $0.25 $1.25 0 — 37.4 — —
Claude 3 Opus Anthropic 2024-03-04 Proprietary unknown — — 200K $15.00 $75.00 0 — 48.9 — —
StarCoder 2 15B BigCode 2024-02-28 Source-available dense 15B — 16K — — — — — 46.3 gguf awq mlx bnb
Gemini 1.5 Pro Google DeepMind 2024-02-15 Proprietary unknown — — 2.1M $0.00 $0.00 0 — 58.9 — —
Mixtral 8x7B Instruct v0.1 Mistral AI 2023-12-11 Open moe 46.7B 12.9B 33K — — — — — — gguf awq gptq mlx fp8 bnb
Yi-34B-Chat 01.AI 2023-11-23 Open weights dense 34B — 4K — — — — — — gguf awq gptq
Mistral 7B v0.1 Mistral AI 2023-09-27 Open dense 7B — — — — — — — — gguf awq gptq exl2 mlx fp8 bnb
Falcon 180B Chat Technology Innovation Institute 2023-09-04 Source-available dense 180B — 2K — — — — — — gguf awq gptq
Llama 2 70B Chat Meta 2023-07-18 Source-available dense 70B — — — — — — — — gguf awq gptq fp8
GPT-4 OpenAI — Proprietary unknown — — — $30.00 $60.00 28.6 — — — —
GPT-3.5 Turbo OpenAI — Proprietary unknown — — — $0.50 $1.50 91.9 — 29.7 — —

Cost and speed via Artificial Analysis ↗; benchmark scores from each model's primary source (model card or paper). Click any row name for the full record.