A selection of open-weight and hosted models, ordered by announcement date. When a source gives only a month, it appears after dated entries in that month, without an invented day. Each entry links to the provider's source. This is editorial coverage, with no blended ranking or inferred adoption numbers.
15 entries · Review dates and sources on every entry · See the change log
Showing 15 of 15 models.
Save this search
Stores the current filters and comparison selection in this browser. Open it later from saved research.
Compare models
Choose two to four models below. Share this page URL to share your selection. Access and terms reflect each entry’s review date and may have changed. Family entries can have different access paths for each variant; read their restrictions.
0 selected. Choose 2 more to compare; up to four models.
Google describes Gemini 4 Carbon as a model for sustained engineering and cybersecurity tasks. It plans initial access for selected cyber defenders, with broader access to follow.
Reviewed tasks
Coding, Research
Reviewed modalities
Unknown
Deployment paths
Unknown
Focus / inputs
Engineering and cybersecurity; input and output contract not specified
Licence category
Unknown
Licence / terms
Public weight availability and licence terms are not established by the announcement.
Google describes Genie 4 as a model for interactive simulations with native audio. Rollout has begun for limited trusted testers; Project Genie and API access are planned.
Reviewed tasks
Unknown
Reviewed modalities
Audio
Deployment paths
Unknown
Focus / inputs
Interactive simulations with generated audio; input contract not specified
Licence category
Unknown
Licence / terms
Public weight availability and licence terms are not established by the announcement.
A model for visual reasoning, computer use and writing. OpenAI positions Terra for frequent, well-defined work such as reading documents and operating software.
A speech model family and voice-agent SDK combining transcription, speech generation and tool calls. Separate models handle listening and speaking so teams can choose their reasoning model.
Reviewed tasks
Speech
Reviewed modalities
Text, Audio
Deployment paths
Hosted API, Vendor-arranged deployment
Focus / inputs
Speech recognition and speech generation
Licence category
Unknown
Licence / terms
Provider terms; self-deployment terms require confirmation with Mistral.
Variants
4; the rows below are for Voxtral Mini Transcribe, the most fully reviewed
Muse Wonder and Muse Wonder Full are designed for tasks that span multiple steps, with memory across conversations. The two variants have different access paths.
Reviewed tasks
Coding, Knowledge work
Reviewed modalities
Unknown
Deployment paths
Hosted API, Provider app
Focus / inputs
Agentic coding and professional tasks
Licence category
Provider terms
Licence / terms
Proprietary; provider terms apply.
Variants
2; the rows below are for Muse Wonder, the most fully reviewed
A smaller Claude model for classification, summaries, routing and subagent work. Haiku now supports adjustable effort and adaptive thinking for tasks with different latency budgets.
Reviewed tasks
Classification and routing, Writing, Coding, Computer use
A compact model that maps different media into a shared embedding space for local search and retrieval. Text, vision and audio components can be selected to fit the application.
Mistral released Large 4 weights for coding, visual reasoning and agent workflows. Its Mistral Ecosystem Open-Weight licence requires a separate commercial licence for businesses operating Model-as-a-Service offerings.
Reviewed tasks
Coding, Knowledge work, Visual reasoning
Reviewed modalities
Text, Images
Deployment paths
Hosted API, Downloadable weights
Focus / inputs
Multimodal reasoning and coding
Licence category
Unknown
Licence / terms
Mistral Ecosystem Open-Weight, a modified MIT licence; commercial Model-as-a-Service businesses require a separate Mistral licence.
Variants
1
Context window
1M tokens
Base input price
0.7 USD/MTok (Standard) Mistral pricing lists 0.5 USD/MTok (Mistral Large (no version stated)), checked .
Base output price
2.1 USD/MTok (Standard) Mistral pricing lists 1.5 USD/MTok (Mistral Large (no version stated)), checked .
A Gemini model designed for long tasks in software engineering and professional work. Google introduced it with a phased rollout, then expanded Gemini app access in October.
Reviewed tasks
Coding, Knowledge work, Research, Writing
Reviewed modalities
Unknown
Deployment paths
Provider app
Focus / inputs
Reasoning, coding and professional tasks
Licence category
Provider terms
Licence / terms
Proprietary; provider terms apply.
Variants
1
Context window
Unified 1M context window
Base input price
2 USD/MTok (Standard), until From : 4 USD/MTok (Standard), checked .
Base output price
10 USD/MTok (Standard), until From : 20 USD/MTok (Standard), checked .
A general model for coding, computer use and professional work. OpenAI positions Sol between its flagship and smaller models, with configurable reasoning effort.
A Claude model for scoped coding tasks and everyday documents, slides and spreadsheets. Anthropic presents it as a faster complement to Opus for iterative work.
A Claude model for demanding work across large codebases and long tasks. Its release changes thinking behavior and integration requirements, which developers should check before migrating.
Two GPT-6 models aimed at different workload sizes and cost budgets. Sol handles demanding everyday tasks, while Luna is the smaller option for faster iteration and larger volumes.
Reviewed tasks
Coding, Computer use, Knowledge work
Reviewed modalities
Unknown
Deployment paths
Hosted API, Provider app
Focus / inputs
Reasoning, coding and computer use
Licence category
Provider terms
Licence / terms
Proprietary; provider terms apply.
Variants
2; the rows below are for GPT-6 Sol, the most fully reviewed
A GPT-6 model for difficult software, research and professional tasks that span multiple steps. It combines reasoning with browser and computer interaction; tools and execution environments determine what an application can do.
Two access configurations of the same underlying model. Fable is generally available; Mythos uses different safeguards for approved specialist work. The official announcement gives a month without an exact day.
Reviewed tasks
Coding, Knowledge work, Research
Reviewed modalities
Text, Images
Deployment paths
Hosted API
Focus / inputs
Fable: text and images → text. Mythos input support is not independently documented here.
Licence category
Provider terms
Licence / terms
Proprietary; provider terms and access restrictions apply.
Variants
2; the rows below are for Claude Fable 5.1, the most fully reviewed
Rising uses Google Analytics only if you accept. It sets cookies and sends page activity to Google. The rest of the site works the same if you refuse. Read the privacy page.
