LLM Models

Google DeepMind's workhorse Gemini model, released July 2026 alongside 3.5 Flash-Lite. Gemini 3.6 Flash improves on 3.5 Flash across coding, knowledge work, and multimodal tasks while using up to 17% fewer output tokens on the Artificial Analysis Index, making it cheaper to run than its predecessor.

Anthropic's model for complex agentic coding and enterprise work, released July 2026. Opus 5 comes close to the frontier intelligence of Claude Fable 5 at half the price, and delivers substantially improved performance at the same $5/$25 pricing as its predecessor Opus 4.8. It has a 1M-token context window, 128K max output, and adaptive thinking with a five-level effort setting.

Google's image generation and editing model (preview), part of the Gemini ecosystem, producing high-fidelity images with strong prompt adherence and editing controls.

Google DeepMind's video generation model, producing high-quality video with audio from text and image prompts. Available in preview through the Gemini API.

Alibaba's large-scale proprietary MoE model (January 2025), pretrained on over 20 trillion tokens and post-trained with SFT and RLHF. Served via the Alibaba Cloud API, Qwen2.5-Max competes with leading frontier models on knowledge, reasoning, and coding benchmarks.

Alibaba's open-weight 32B reasoning model (March 2025, Apache 2.0), which uses reinforcement learning to reach reasoning performance competitive with much larger models and to beat o1 on some benchmarks, at a size that runs on a single high-end GPU.

Alibaba's most powerful open-weight coding model (July 2025), a 480B-parameter mixture-of-experts with 35B active parameters built for agentic coding. It targets repository-scale software tasks with a very long context window and strong tool-use and code-generation performance.

Alibaba's flagship open-weight Qwen3 model (April 2025), a 235B-parameter mixture-of-experts that activates 22B parameters per token. Trained on ~36T tokens across 119 languages under Apache 2.0, it supports seamless switching between a thinking mode for hard reasoning and a fast non-thinking mode, and is competitive with DeepSeek-R1, o1, o3-mini, and Gemini 2.5 Pro.

Alibaba's flagship proprietary Qwen model (May 2026) with a 1M-token context window, native extended-thinking mode, and benchmark wins on SWE-Pro and Terminal-Bench. API-only via DashScope (no open weights).

Alibaba's February 2026 Qwen update: a 397B-total / 17B-active sparse mixture-of-experts model with a 1M-token native context window and coverage of 201 languages. Open-weight tiers are released under the Apache 2.0 license.

Mistral's April 2026 mid-tier model: a 128B dense transformer with a 256K-token context window and strong coding performance (77.6% on SWE-bench Verified), released as open weights under a modified MIT license.

Mistral's December 2025 open-weights flagship: a 675B-total / ~41B-active sparse mixture-of-experts model with native vision, a 256K-token context window, and strong multilingual support - released under the permissive Apache 2.0 license.