AI models ranked
Curated rankings of frontier and open-source AI models for developers. Scores are calibrated against the current frontier so older generations do not look artificially competitive.
65 models - page 1 of 6
Google's October 2026 flagship reasoning model for complex software engineering, autonomous agents, math, science, and multimodal analysis. Its 1,048,576-token window is shared across input, thinking, and output; background long decode can use the remaining window beyond the default 65,536-token synchronous output limit. The listed $2 / $10 token rates are introductory through November 30, 2026.
OpenAI's September 2026 premium model for complex reasoning, software engineering, research, computer use, and professional tools. Its 1.05M context window and 128K maximum output support long multistep workflows, while the $10 / $50 token rates favor high-value tasks.
Anthropic's September 2026 model for long-running agentic coding and knowledge work. It pairs a 1M context window and 128K maximum output with always-on adaptive thinking, moderate comparative latency, and $4 / $20 token rates that are lower than Opus 5.
Anthropic's September 2026 premium model for demanding reasoning, long-horizon agentic coding, research, and professional work. Published evaluations place it effectively level with Astra on Terminal-Bench 4.0, while its $0.25 cache-read rate improves repeated-context agent economics.
xAI's September 2026 model for coding, agentic tasks, and knowledge work. It offers configurable reasoning, a 500K context window, search and code-execution tools, and base rates of $2 / $6 per million tokens; prompts above 200K input tokens use xAI's higher long-context tier.
Anthropic's July 2026 model for complex agentic coding and enterprise work. Opus 5.5 succeeds it with lower token and cache-read prices, while retaining the same 1M context window and 128K maximum output.
The same underlying model as Fable 5.1 with reduced cybersecurity and biology safeguards for vetted Project Glasswing participants. Its higher specialist ceiling is reflected in coding, while limited availability lowers its general Frontier Fit rather than its capability grades.
OpenAI's July 2026 high-capability GPT-5.6 tier for complex professional work, deep reasoning, long-horizon coding agents, and demanding tool workflows. It supports configurable reasoning through max effort at lower token rates than GPT-6 Astra.
OpenAI's April 2026 GPT-5 model for agentic coding, professional knowledge work, data analysis, computer use, and long-running tool workflows. Its 1.05M context and broad Responses API tool support remain useful, while the later GPT-5.6 and GPT-6 families occupy OpenAI's premium catalog tiers.
Google's May 2026 Gemini production model for coding, agentic loops, grounding, tool use, and multimodal inputs. It combines a 1M-token context window with lower token rates than Gemini 3.1 Pro Preview in this catalog.
xAI's May 2026 general-purpose model with configurable reasoning, agentic tool calling, image input, and a 1M-token context window. Grok 4.7 succeeds it for coding and knowledge work; use search tools when current information matters.
DeepSeek's V4 Pro open-weight model for agentic coding, math, STEM reasoning, and 1M-context workflows. It supports OpenAI-compatible and Anthropic-compatible APIs with thinking and non-thinking modes.