46 families, 55 access platforms. Search first; open a card for exact reviewed variants and comparison.
46 model families shown
MiniMaxMixed
MiniMax H3
MiniMax H3 · hosted · MiniMax H3-Base
Video generation with synchronized audio; hosted and downloadable base configurations differ.
Consider for
Video and sound production, reference-driven scenes and local workflow experiments.
VideoAudioOpen weightsTextImage
Family overview reviewed 2026-09-08
2 reviewed variants · access & comparison
MiniMax H3 · hosted
Available
MiniMax-H3
Video. Generate video with synchronized sound from text and references. 4–15 seconds. Full hosted system supports output up to 2K; that is not the downloadable base checkpoint’s resolution.
Video. Build a local video-and-audio workflow with downloadable weights. 768p base output. FL2VA uses text and first/last frames; Ref2VA accepts multimodal references. Context-IR and Regenerate-2K are not released. Check the community license and runtime requirements.
Hosted video generation and editing with multimodal references.
Consider for
Reference-driven video production and longer takes.
VideoTextImageAudio
Family overview reviewed 2026-09-08
1 reviewed variants · access & comparison
Seedance 2.5
Available
dreamina-seedance-2-5-260628
Video. Reference-led scenes, longer takes, and targeted video editing. Up to 30 seconds with multi-round extension. The API ID shown belongs to BytePlus LAS. Consumer rollout, account access and API billing are separate.
Speech synthesis with separate HD and Turbo tiers.
Consider for
Narration, voiceovers and spoken application output.
AudioText
Family overview reviewed 2026-09-08
2 reviewed variants · access & comparison
MiniMax Speech 2.8 HD
Available
speech-2.8-hd
Speech. Try for polished narration and multilingual voiceovers. Text-to-speech with streaming, pronunciation controls and voice selection. Compare HD and Turbo on the same script.
Speech. Try for interactive or repeated speech generation. A separate speech tier. Measure response time, pronunciation and delivered-audio cost on your workload.
Streaming text-to-speech for conversational applications.
Consider for
Responsive voice interfaces and spoken assistants.
AudioText
Family overview reviewed 2026-09-08
1 reviewed variants · access & comparison
Cartesia Sonic 3.6
Available
sonic-3.6
Speech. Test conversational speech and responsive spoken interfaces. Generally available August 27, 2026; 44 languages. Pin sonic-3.6-2026-08-27 for a reproducible model snapshot.
Downloadable song generation from descriptions and lyrics.
Consider for
Music demos, song sketches and local audio workflows.
AudioOpen weightsText
Family overview reviewed 2026-09-08
1 reviewed variants · access & comparison
MiniMax Music 3
Downloadable
MiniMaxAI/MiniMax-Music3
Music. Turn a musical brief and lyrics into a complete song. Up to five minutes, 32 kHz stereo output. Review the model’s license and official workflow requirements before production use.
Sol, Terra, and Luna offer different capability and cost tiers for professional work. Sol currently has a time-limited promotional price.
Consider for
Professional reasoning, coding, computer use, agents, and scaled extraction or routing.
FrontierEfficientTextImageTools
Family overview reviewed 2026-09-07
3 reviewed variants · access & comparison
GPT-5.6 Sol
Available
gpt-5.6-sol
1.05M total; 922K input / 128K output. Reasoning, coding, and tool use. Promotional rate confirmed at least through November 21; later rate is not established.
Still-available, previous-generation, coding, chat, research, and reasoning models. Useful when an existing integration needs a pinned behavior or specialized endpoint.
Consider for
Maintaining existing systems, deep research, Codex workflows, and compatibility-sensitive deployments.
FrontierEfficientSpecialistTextImageTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
GPT-Realtime 2.1 · Realtime mini · Live Transcribe · Transcribe · Audio 1.5 · Whisper
Separate families for low-latency voice conversations, transcription, translation, speech generation, and audio-in/audio-out applications.
Consider for
Voice agents, meetings, captions, call automation, and live translation.
AudioSpecialistTextTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
GPT Image 2 · legacy image variants · Sora 2 / Pro (deprecated)
GPT Image 2 is the current API image choice. Sora 2, Sora 2 Pro, and the Videos API have a scheduled September 24, 2026 shutdown. Legacy image variants have separate retirement dates.
Consider for
Image creation and editing with GPT Image 2. Review lifecycle notices before maintaining a legacy image or video integration.
ImageVideoAudioText
Family overview reviewed 2026-09-07
7 reviewed variants · access & comparison
GPT Image 2
Available
gpt-image-2
Image generation and edits through the Image API; compatible Responses image tool workflows. Account verification may be required.
Fable is publicly offered, subject to plan and API billing. Mythos requires approved, limited access. Exact cloud availability and hosting differ by variant.
Consider for
Maximum-depth agents, long-horizon enterprise work, and approved defensive security programs.
FrontierSpecialistTextImageTools
Family overview reviewed 2026-09-07
4 reviewed variants · access & comparison
Claude Fable 5.1
Available
claude-fable-5-1
1M total / 128K output. Long-running agents and difficult reasoning. Account access and usage billing apply.
Invitation-only. No ordinary Claude chat-picker or Claude Platform on AWS availability established. Confirm Google Cloud access directly; precise provider-side variant status is unverified here.
Invitation-only. No ordinary Claude chat-picker or Claude Platform on AWS availability established. Confirm Google Cloud access directly; precise provider-side variant status is unverified here.
Opus 5 is a premium general and agentic model with a one-million-token context window. Check each earlier model separately for lifecycle and availability.
Consider for
Complex coding, deep reasoning, document work, code review, and sustained agents.
FrontierTextImageTools
Family overview reviewed 2026-09-07
1 reviewed variants · access & comparison
Claude Opus 5
Available
claude-opus-5
1M total / 128K output. Complex coding, analysis, and sustained agents. Account access and usage billing apply.
Sonnet 5 targets daily coding and agents; its $2/$10 base API price is permanent after the planned increase was cancelled. Haiku 4.5 is a smaller, fast tier.
Consider for
Daily coding and agents, customer workflows, document analysis, triage, and high-volume production.
FrontierEfficientTextImageTools
Family overview reviewed 2026-09-07
2 reviewed variants · access & comparison
Claude Sonnet 5
Available
claude-sonnet-5
1M total / 128K output. Daily coding and agent workflows. Account access and usage billing apply.
Gemini 3.1 Pro Preview · managed research variants
Gemini 3.1 Pro Preview supports large multimodal inputs. Prompts above 200K tokens have higher input and output rates; managed research products have separate terms.
Consider for
Difficult multimodal reasoning, grounded research, large documents, coding, and agentic workflows.
FrontierSpecialistTextImageAudioVideoTools
Family overview reviewed 2026-09-07
1 reviewed variants · access & comparison
Gemini 3.1 Pro Preview
Preview
gemini-3.1-pro-preview
1,048,576 input / 65,536 output. Difficult multimodal reasoning and large documents. Account access and usage billing apply.
Google’s efficient multimodal family. Gemini 3.8 Flash is the latest Flash entry, with promotional token rates through December 31, 2026.
Consider for
Multimodal extraction, fast agents, video understanding, classification, and structured output.
FrontierEfficientTextImageAudioVideoTools
Family overview reviewed 2026-09-07
2 reviewed variants · access & comparison
Gemini 3.8 Flash
Available
gemini-3.8-flash
1,048,576 input / 65,536 output. Fast multimodal work and grounded research. 50% promotional rate through December 31; $1.50 / $7.50 from January 1, 2027.
Voice applications, translation, UI automation, and managed multi-step work.
AudioSpecialistTextImageTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Nano Banana 2 · 2 Lite · Pro · Veo 3.1 · Lyria 3 Pro · Lyria 3 Clip · Lyria RealTime
Google’s native image editing, video generation, and music-generation families across consumer and developer surfaces.
Consider for
Image creation and editing, video with audio, music clips, and realtime generative music.
ImageVideoAudioText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Image creation/editing and image-to-video models distributed through Grok and xAI’s developer platform.
Consider for
Rapid social imagery, image edits, and short image-to-video clips.
ImageVideoText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Medium 3.5 · Small 4 · Large 3 · Ministral 3 (3B/8B/14B)
A mix of hosted frontier models and permissively licensed open-weight models spanning server, workstation, and compact deployment.
Consider for
European enterprise deployment, agents, coding, multimodal work, and local inference.
FrontierEfficientOpen weightsTextImageTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Purpose-built models for code completion, transcription, speech generation, OCR, moderation, theorem proving, and embeddings.
Consider for
Focused production tasks where a specialized model is cheaper and easier to evaluate than a general LLM.
AudioRetrievalSpecialistTextImageCode
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Command A+ · A · A Reasoning · A Vision · A Translate · R7B · Aya
Enterprise generation, RAG, agents, reasoning, translation, vision, and multilingual open research models.
Consider for
Enterprise retrieval workflows, private deployment, multilingual applications, and tool-using agents.
FrontierEfficientOpen weightsTextImageTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Dedicated retrieval and speech models for semantic indexing, relevance ranking, and transcription.
Consider for
Search, RAG, enterprise knowledge systems, multilingual retrieval, and audio ingestion.
RetrievalAudioSpecialistTextImage
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Nova 2 Lite · 2 Pro · Premier · Lite · Micro · Pro
Amazon’s native Bedrock family for text and multimodal understanding, reasoning, built-in tools, and multiple cost/capability tiers.
Consider for
AWS-native applications, document and video understanding, agents, and enterprise controls.
FrontierEfficientTextImageVideoTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Nova 2 Sonic · Nova Sonic · Canvas · Reel · Multimodal Embeddings
Speech-to-speech, image, video, and multimodal embedding models that stay inside the Bedrock service model.
Consider for
AWS voice agents, image/video creation, and multimodal retrieval.
AudioImageVideoRetrievalText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
A one-million-context open-weight family with thinking and non-thinking modes, tool calls, structured output, and inexpensive first-party API access.
Consider for
Reasoning, agentic coding, large-context work, and self-hosted experimentation.
FrontierEfficientOpen weightsTextToolsCode
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Planning estimate: 2–4 GiB for model + runtime at 4K context; 16 GiB system/unified memory is a more comfortable starting point. Not a benchmark or full-GPU-fit guarantee.
1.28 GB download · candidate from 8 GiB memory (estimate)
Planning estimate: 4–6 GiB for model + runtime at 4K context; 16 GiB system/unified memory is a more comfortable starting point. Not a benchmark or full-GPU-fit guarantee.
2.5 GB download · candidate from 8 GiB memory (estimate)
Planning estimate: 6–9 GiB for model + runtime at 4K context; 16 GiB system/unified memory is a more comfortable starting point. Not a benchmark or full-GPU-fit guarantee.
4.68 GB download · candidate from 16 GiB memory (estimate)
Planning estimate: 11–15 GiB for model + runtime at 4K context; 24 GiB system/unified memory is a more comfortable starting point. Not a benchmark or full-GPU-fit guarantee.
9 GB download · candidate from 24 GiB memory (estimate)
Planning estimate: 22–28 GiB for model + runtime at 4K context; 48 GiB system/unified memory is a more comfortable starting point. Not a benchmark or full-GPU-fit guarantee.
19.9 GB download · candidate from 32 GiB memory (estimate)
Qwen Image 2.0/Pro · Wan 2.7 image and video variants
Alibaba’s image generation/editing and open video model families, with community ComfyUI workflows and multiple hosted APIs.
Consider for
Local or hosted image generation, text rendering, image-to-video, and controllable video pipelines.
ImageVideoOpen weightsText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Long-context general and coding models from Moonshot AI, distributed through the Kimi platform and several open-model hosts.
Consider for
Large-context work, agents, software engineering, and open-weight deployment.
FrontierOpen weightsTextImageToolsCode
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
A frontier Chinese and multilingual family with coding, reasoning, tool use, vision, hosted APIs, and open releases.
Consider for
Coding agents, multilingual reasoning, long context, and local or hosted experiments.
FrontierOpen weightsTextImageToolsCode
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Language models for long-context applications and agents. Creative models have separate H3, Speech and Music records.
Consider for
Cost-sensitive agents and multilingual language applications.
FrontierTextImageTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Meta’s open-weight ecosystem spans multimodal mixture-of-experts models, compact edge models, very large text models, and safety classifiers.
Consider for
Self-hosting, customization, broad platform choice, local chat, and enterprise open-model stacks.
Open weightsSpecialistTextImageTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Google’s open model family now ranges from mobile-scale multimodal models through 31B server models, plus safety and vision-language specialists.
Consider for
On-device AI, laptops, local multimodal work, research, and commercially usable open deployment.
Open weightsEfficientSpecialistTextImageAudioVideoTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Nemotron 3 Nano · Super 120B A12B · Ultra 550B A55B · safety, ASR, and domain variants
NVIDIA’s production-oriented open family, offered as downloadable weights and optimized NIM containers across a wide range of NVIDIA hardware.
Consider for
DGX Spark, NVIDIA enterprise deployment, local agents, domain adaptation, safety, and speech.
Open weightsAudioSpecialistTextToolsCode
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
OpenAI’s open-weight reasoning family, including a large model, a compact model, and a safety-focused variant.
Consider for
Self-hosted reasoning, local agents, coding, and high-speed open-model inference.
Open weightsEfficientSpecialistTextToolsCode
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Phi-4 · Phi-4 mini · multimodal, reasoning, and local variants
Microsoft’s small-language-model family, optimized for compact deployment, ONNX, local runtimes, and Microsoft Foundry.
Consider for
On-device assistants, constrained hardware, task-specific fine-tuning, and offline applications.
Open weightsEfficientTextImageAudioTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Granite 4 · code, vision, embedding, guardian, and time-series families
IBM’s enterprise open family spans general language, code, documents, embeddings, safety, and forecasting.
Consider for
Enterprise customization, document workflows, governed open-model stacks, and specialized business tasks.
Open weightsRetrievalSpecialistTextImageCodeTime series
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
AI21’s hybrid architecture family for long context, tool use, structured output, and enterprise deployment.
Consider for
Long documents, enterprise agents, RAG, and workloads distributed through AWS.
FrontierEfficientOpen weightsTextTools
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Compact foundation models designed for memory efficiency and deployment from edge devices through servers.
Consider for
Edge and local inference, efficient agents, and alternative model architecture experiments.
Open weightsEfficientTextImageAudio
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
A broad image-generation and editing family with hosted premium tiers plus downloadable variants and strong ComfyUI support.
Consider for
High-quality images, editing, typography, product visuals, LoRAs, and repeatable local workflows.
ImageOpen weightsText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Open and hosted creative model families for images, sound effects, music, editing, and video experimentation.
Consider for
Local creative pipelines, controllable generation, fine-tuning, sound design, and image editing.
ImageVideoAudioOpen weightsText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Two major open video ecosystems with text-to-video, image-to-video, reference-to-video, and native audio-video workflows.
Consider for
Local storyboards, controllable shots, research, and ComfyUI-based video production.
VideoAudioOpen weightsImageText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Runway Gen and Aleph families · Luma Ray and Photon families
Commercial creative suites with proprietary image, video, editing, camera, and transformation models exposed through products and APIs.
Consider for
Professional creative iteration, video transformation, visual effects, and cloud-first production.
ImageVideoText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Adobe Firefly model family · Ideogram 4 and prior generations
Commercial creative models centered on design workflows, text rendering, image editing, video, and Adobe application integration.
Consider for
Brand and design work, typography-heavy images, enterprise creative controls, and production editing.
ImageVideoAudioText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
Speech. A speed-focused speech option for interactive applications. Test dates, currencies and phone numbers: text normalization differs from other models.
Consumer-first music creation platforms for full songs, lyrics, styles, extensions, remixes, and stem-oriented workflows.
Consider for
Song ideation, demos, variations, and rapid music experiments where product access matters more than model IDs.
AudioText
Family overview reviewed 2026-08-09 · Review due
Variant and platform coverage
Exact variant-by-platform availability has not been independently verified for this family. The links below identify places to investigate; they do not establish that every listed variant runs on every platform.
A family overview groups related releases. Exact availability is established only where a reviewed variant lists a platform and source. Older summaries retain their original dates; memory estimates are clearly labeled. Read how we research the guide.