Choose a family, then compare exact versions.
Search by model, version, or task.
A broad proprietary model family with a mature API and tool ecosystem.
3 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
gpt-5.6-sol | Highest capability | |
gpt-5.6-terra | Quality and cost balance | |
gpt-5.6-luna | Speed and volume |
Covers general, coding, vision, and tool-based workflows.
Closed weights and provider-controlled deployment.
A proprietary family positioned for analysis, writing, coding, and long-context work.
4 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
claude-fable-5 | Highest capability | |
claude-sonnet-5 | Quality and cost balance | |
claude-opus-4-8 | Coding and agents | |
claude-haiku-4-5-20251001 | Speed and volume |
Well suited to document analysis and sustained written work.
Closed weights and API-dependent access.
Google's proprietary family for text, image, audio, video, and tool-connected workflows.
6 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
gemini-3.5-flash | Highest capability | |
gemini-3.1-pro-previewPreview | Deep reasoning | |
gemini-3.1-flash-lite | Speed and volume | |
gemini-2.5-proEnds 2026-10-16 | Earlier stable release | |
gemini-2.5-flashEnds 2026-10-16 | Earlier stable release | |
gemini-2.5-flash-liteEnds 2026-10-16 | Earlier stable release |
Native multimodal options across several input types.
Closed weights for flagship hosted models.
An open-weight family designed for self-hosting, adaptation, and a broad community ecosystem.
3 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
meta-llama/Llama-4-Maverick-17B-128E-Instruct | Large open model | |
meta-llama/Llama-4-Scout-17B-16E-Instruct | Efficient open model | |
meta-llama/Llama-3.3-70B-Instruct | Earlier stable release |
Greater control over deployment, data path, and customization.
Production serving requires hardware, security, and operations work.
A reasoning- and coding-oriented family available through APIs and selected open-weight releases.
2 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
deepseek-v4-proPreview | Highest capability | |
deepseek-v4-flashPreview | Speed and volume |
Offers explicit reasoning and coding-oriented model options.
Names and compatibility aliases can change quickly.
A multilingual family with Chinese-focused, coding, multimodal, API, and open deployment options.
12 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
qwen3.7-max-2026-06-08 | Highest capability | |
qwen3.7-plus-2026-05-26 | Quality and cost balance | |
qwen3.6-plus-2026-04-02 | Earlier stable release | |
qwen3-coder-plus-2025-09-23 | Coding and agents | |
Qwen/Qwen3-235B-A22B | Large open model | |
Qwen/Qwen3-30B-A3B | Efficient open model | |
Qwen/Qwen3-32B | Large open model | |
Qwen/Qwen3-14B | Efficient open model | |
Qwen/Qwen3-8B | Efficient open model | |
Qwen/Qwen3-4B | Local and edge | |
Qwen/Qwen3-1.7B | Local and edge | |
Qwen/Qwen3-0.6B | Local and edge |
Broad Chinese and multilingual model coverage.
The family spans many specialized versions that are not interchangeable.
A family mixing managed, open-weight, efficient, coding, and specialist models.
6 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
mistral-medium-3-5 | Quality and cost balance | |
mistral-small-2603 | Speed and volume | |
mistral-large-2512 | Large open model | |
ministral-14b-2512 | Efficient open model | |
ministral-8b-2512 | Local and edge | |
ministral-3b-2512 | Local and edge |
Multiple sizes and deployment paths support efficiency trade-offs.
No single Mistral model represents the whole family.
xAI's proprietary family for reasoning, tool use, multimodal input, and live-data workflows.
5 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
grok-4.5 | Highest capability | |
grok-4.3 | Quality and cost balance | |
grok-4.20-0309-reasoning | Deep reasoning | |
grok-4.20-0309-non-reasoning | Speed and volume | |
grok-build-0.1Preview | Coding and agents |
Managed models include tool and real-time information paths.
Closed weights and provider-controlled data/tool access.
Cohere's enterprise-oriented generation family for RAG, agents, tool use, and multilingual work.
8 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
command-a-plus-05-2026 | Highest capability | |
command-a-03-2025 | Enterprise and RAG | |
command-a-reasoning-08-2025 | Deep reasoning | |
command-a-vision-07-2025 | Multimodal work | |
command-a-translate-08-2025 | Translation | |
command-r7b-12-2024 | Speed and volume | |
command-r-plus-08-2024 | Earlier stable release | |
command-r-08-2024 | Earlier stable release |
Product line is explicitly designed around retrieval and enterprise workflows.
RAG quality depends heavily on retrieval, documents, and citations configuration.
Google's lightweight open-model family for local experimentation, adaptation, and selected multimodal tasks.
5 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
google/gemma-4-E2B-it | Local and edge | |
google/gemma-4-E4B-it | Local and edge | |
google/gemma-4-12B-it | Efficient open model | |
google/gemma-4-26B-A4B-it | Efficient open model | |
google/gemma-4-31B-it | Large open model |
Downloadable models support private and customizable deployments.
Smaller footprints trade capability for resource efficiency.
Microsoft's small open-model family aimed at efficient local, edge, and constrained deployments.
5 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
microsoft/phi-4 | Efficient open model | |
microsoft/Phi-4-mini-instruct | Local and edge | |
microsoft/Phi-4-multimodal-instruct | Multimodal work | |
microsoft/Phi-4-reasoning | Deep reasoning | |
microsoft/Phi-4-mini-reasoning | Local and edge |
Small footprints can suit on-device and low-resource scenarios.
Compact size creates task and capacity trade-offs.
Moonshot AI's hosted family positioned for long-context, Chinese, reasoning, and agent workflows.
11 versions · Reviewed Jul 2026
| Select for comparison | Model / version | Best fit |
|---|---|---|
kimi-k3 | Highest capability | |
kimi-k2.7-code | Coding and agents | |
kimi-k2.7-code-highspeed | Speed and volume | |
kimi-k2.6 | Multimodal work | |
kimi-k2.5Ends 2026-08-31 | Earlier stable release | |
moonshot-v1-8kEnds 2026-08-31 | Earlier stable release | |
moonshot-v1-32kEnds 2026-08-31 | Earlier stable release | |
moonshot-v1-128kEnds 2026-08-31 | Earlier stable release | |
moonshot-v1-8k-vision-previewEnds 2026-08-31 | Multimodal work | |
moonshot-v1-32k-vision-previewEnds 2026-08-31 | Multimodal work | |
moonshot-v1-128k-vision-previewEnds 2026-08-31 | Multimodal work |
Hosted APIs emphasize long documents and Chinese-language use.
Closed hosted access limits deployment control.
OpenGPTEvidence-based model decisions