Model database

Google

Gemini 4 Argon is Google's September 30, 2026 frontier model. The launch post says it is rolling out only to trusted cyber defenders in the Fairwind Program, with broader developer, enterprise, and consumer access planned later, starting with the paid Gemini API and Google AI Ultra. Google has not published a Gemini API model ID, and the public models list does not include Argon. The same post sets an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input at 95% off the input price, then $4 and $20 per million after the introductory period. It does not give the end date of that period or an input context window. Output tokens are listed at 1 million, up from a previous 64,000. Vendor-reported scores on the post include 77.9% on DeepSWE v1.1, 51.3% on AutomationBench, 91.7% on LVBench, and a first-place tie at 68% on CWE-bench v1.

Google

Gemini Nano Banana 2.1 is Google's generally available image model, released October 6, 2026 on the Gemini API as gemini-nano-banana-2.1. It updates Nano Banana 2 (gemini-3.1-flash-image), which the changelog deprecates with a shutdown on October 29, 2026. The model page lists a 131,072-token input limit and a 32,768-token output limit, with text, image, video, and PDF inputs and image plus text outputs. Resolutions are 1K by default, plus 2K and 4K, including wide ratios 1:4, 4:1, 1:8, and 8:1. It accepts up to 14 reference images, with up to 4 characters and up to 10 objects, and thinking levels minimal, medium (the default), and high. Google Search grounding and the Batch API are supported. The Gemini pricing page fetched October 7, 2026 lists rates for Nano Banana 2 and does not list a separate dollar rate for gemini-nano-banana-2.1.

Google

EmbeddingGemma 2 is Google's open embedding model, published October 6, 2026 under the Apache 2.0 license. It is not a generative Gemini API model. The launch post says it has 740 million parameters: a 270 million text encoder, an optional 170 million vision encoder, and a 300 million audio encoder. Embeddings are 768 dimensions and can be truncated to 512, 256, or 128. Context is 8,000 tokens, four times EmbeddingGemma 1. The post says it can process up to 5.5 minutes of audio, 29 images, or 58 video frames, and that the text-only configuration can run at 270 million parameters. Weights are on Hugging Face and Kaggle. Google says a Model Garden listing is coming soon. A vendor score on the post moves MTEB Code from 68.76 to 78.68.

Mistral AI

Mistral Large 4 is Mistral's public-preview flagship, announced October 6, 2026. The launch post describes a 1-trillion-parameter model with 49 billion active parameters, trained on 3,800 NVIDIA Grace Blackwell GPUs, and says the weights will be released at the end of the month. The same-day model card lists 1.05 trillion total parameters, 52 billion active parameters, a 1.6 billion vision encoder, and a 1 million token context window. The pricing page lists a sale price of $0.68 input, $0.07 cached input, and $2.09 output per million tokens, and an original price of $1.36, $0.14, and $4.18. The preview API is on Mistral Studio. The model card renders the model string as mistral-large-4.

Anthropic

Claude Sonnet 5.5 is Anthropic's September 28, 2026 Sonnet model, documented as a clear upgrade over Claude Sonnet 5 that runs more than 30% faster and costs up to 30% less for most work. The Claude API ID is claude-sonnet-5-5. It has a 1M-token context window, 128K synchronous max output, text and image input, text output, adaptive thinking, and default effort high. Standard price is $2 input and $10 output per million tokens.

OpenAI

GPT-6.1 Sol is OpenAI's current Sol model, positioned as near-Astra performance for complex work at a lower cost than Astra. The API ID is gpt-6.1-sol. It accepts text and image input, returns text, and has a 1,050,000-token context window, 128,000 max output, and a knowledge cutoff of April 30, 2026. Standard price is $2 input, $0.10 cached input, $2.50 cache writes, and $10 output per million tokens.

Google

Gemini 3.8 Flash TTS is Google's September 2026 creative and studio text-to-speech model. The model code is gemini-3.8-flash-tts. It takes text input and returns audio, with an 8,192-token input limit, a 16,384-token output limit, and 130 languages. Default unary audio is WAV. Google's Gemini API pricing page does not list a paid rate for this model ID.

ElevenLabs

Eleven v3 is ElevenLabs' documented flagship for human-like, expressive speech. The model ID is eleven_v3. It supports 70+ languages and is used with the Text to Speech API and the Text to Dialogue API. The official models page lists a 5,000-character request limit, about five minutes of audio. Help Center still calls Eleven v3 the latest speech model; Eleven v4 is not on the models page.

Sarvam AI

Sarvam Vision is Sarvam AI's document vision model for Indic and English pages. The documented API model ID is sarvam-vision, a 3B vision-language model covering 23 languages (22 Indian languages plus English). Document AI digitise and extract accept up to 10 pages per PDF and 200 MB per file, at 10 requests per minute. Sarvam's public pricing page lists ₹0.50 per page. The September 24, 2026 post describes Vision 2.1 as a capability update; the docs still use sarvam-vision.

OpenAI

GPT-6 Astra is OpenAI's September 2026 flagship for the hardest end-to-end reasoning, coding, computer use, research, and document work. The API model ID is gpt-6-astra, with a 1,050,000-token context window, 128,000 max output tokens, text and image input, and reasoning.effort from low through max. OpenAI's safety overview (September 3, 2026) states Astra is the first broadly deployed OpenAI model to reach the Critical cybersecurity capability level under the Preparedness Framework.

Anthropic

Claude Fable 5.1 is Anthropic's September 1, 2026 generally available flagship for demanding reasoning and long-horizon agentic coding, research, and document work. The API ID is claude-fable-5-1, with a 1M-token context window, 128K max output, adaptive thinking, and default high effort. Anthropic keeps the $10/$50 input/output list price of Fable 5 while cutting cache reads to $0.25 per million tokens.

Google

Gemini 3.8 Flash is Google's September 2, 2026 workhorse for software engineering, agentic tasks, and multi-step reasoning at Flash speed. The DeepMind model card documents text, image, audio, and video input, a 1M-token context window, 64K output, and adjustable effort levels. Google keeps the same introductory price as 3.7 Flash: $0.75 input and $3.75 output per million tokens through December 31, 2026.

Google

Gemini 3.8 Live is Google's September 15, 2026 default Live API model for low-latency voice agents and real-time dialogue. The stable model code is gemini-3.8-live. It takes text, images, audio, and video, and returns text and audio, with a 131,072-token input limit and 65,536-token output limit. Google positions it for scale and cost efficiency: interleaved reasoning, asynchronous function calling, 97-language mid-conversation switching, and visual grounding while the session keeps talking. Audio sessions are listed at $0.005 per minute input and $0.018 per minute output on the developer Live post.

TypeSafe AI

Jev is TypeSafe AI's first System One model, launched in early access on September 15, 2026. It is not a chatbot: you send unstructured or structured state plus typed questions and get structured answers your code can branch on. The documented model alias is jev-latest on POST https://api.typesafe.ai/v1/systemone. Question primitives are Choice (up to 255 options), Score, and Noul (yes/no probability). TypeSafe lists $0.042 per million input tokens with output tokens unmetered, and 70–500 ms end-to-end on its launch workloads. Training is Reinforcement Learning for Calibrated Decisions (RLCD).

OpenAI

GPT-Live-1 is OpenAI's full-duplex voice model for real-time spoken conversation. The API ID is gpt-live-1. It listens and speaks at the same time, handles interruptions, and delegates reasoning and tool use to a backend agent instead of doing that work itself. OpenAI's system card (July 8, 2026) introduced it as the paid ChatGPT voice default. The API launch post (September 10, 2026) documents Live sessions, native ASR transcripts and response text, keyword biasing, and turn detection. Voice sessions are billed $0.05 per minute, per second. Backend model and tool usage is billed separately.

Meta

Muse Spark 1.3 is Meta's September 2, 2026 hosted model for long-horizon agentic and coding work. The API ID is muse-spark-1.3, with a 1,048,576-token context window, text/image/video/PDF input, and text output. Meta documents it in Muse Code and Meta Model API, with reasoning effort including max on the Standard tier. Official docs still list muse-spark-1.2 as the Muse Code default while recommending 1.3 for new API work.

DeepSeek

DeepSeek-V4.1-Flash is DeepSeek's September 10, 2026 multimodal Mixture-of-Experts model and the smallest member of its new architecture family. The live API ID is deepseek-flash. Official materials describe a 552B-parameter Causal Encoder–Decoder MoE that activates 8B parameters on input and 16B on output, native image-and-text understanding, and a 1-million-token context. DeepSeek says V4.1-Flash outperforms DeepSeek-V4-Pro on performance, cost, speed, and total runtime. Weights are on Hugging Face under MIT.

SpaceXAI

Grok 4.6 is SpaceXAI's frontier model for coding, long-running agents, knowledge work, and interactive visual projects. Official documentation lists text and image input, text output, a 500,000-token context window, configurable low through xhigh reasoning, function calling, web and X search, code execution, and the API model ID grok-4.6.

Z.ai

GLM-5.3-Flash is Z.ai's first natively multimodal GLM-5 model: a 320B-parameter mixture-of-experts checkpoint with 18B active parameters, hybrid sparse and linear attention, and MIT-licensed weights. Z.ai documents API access as glm-5.3-flash, Coding Plan and ZCode availability, and Hugging Face weights for SGLang, vLLM, and TokenSpeed.

Alibaba Qwen

Qwen3.8-Max is Alibaba Qwen's hosted 2.4-trillion-parameter mixture-of-experts flagship for coding, professional work, multimodal analysis, and long-horizon agents. Effective September 5, 2026 (UTC+8), the qwen3.8-max endpoint automatically serves the September 2 snapshot qwen3.8-max-0902 (alias qwen3.8-max-2026-09-02). QwenCloud says 0902 improves coding depth, multi-tool agent delivery, and visual understanding while keeping the 1M context, thinking mode, tools, and unchanged billing.

Microsoft AI

MAI-Image-2.6 is Microsoft AI's September 4, 2026 highest-precision image generation and editing model, in public preview on Microsoft Foundry and MAI Playground. Microsoft says it ranks No. 2 for text-to-image and image editing on Arena and Artificial Analysis as of September 4, with multi-image reference editing, web grounding, dynamic aspect ratios, and up to 1.5K resolution.