AI model leaderboard · LinkWord

OpenAI's flagship GPT-5.6 model. HealthBench Professional 60.5%, High on cybersecurity and biosecurity evals. New API Fast mode: 2.5x faster than Standard at 2x price, no change in intelligence (2026-07-30).

Anthropic's most advanced flagship model with industry-leading reasoning and safety features

Moonshot AI's latest flagship model. 2.8T-parameter MoE (16/896 experts active), 1M context window, native vision. AA Intelligence Index #4, trailing only Fable 5 and GPT-5.6 Sol. Open weights releasing July 27.

Google's Flash-tier workhorse model released Aug 13, 2026 for coding and agents, intro-priced at half of 3.6 Flash's original rate ($0.75/$3.75 per 1M tokens). FrontierCode 1.1: 43.6%, DeepSWE v1.1: 65.3%.

Meta Muse Spark 1.2 multimodal model with 1M context, successor to Muse Spark 1.1

Google's latest Gemini Flash series flagship for coding, knowledge work, and multimodal tasks. 17% less output tokens, DeepSWE 49%, OSWorld-Verified 83.0%. API $1.50/$7.50.

Anthropic Claude mid-range model, balancing performance and cost

Zhipu AI's latest flagship model with strong bilingual understanding and reasoning performance

Fastest, most affordable GPT-5.6 model. API price cut 80% on 2026-07-30 ($0.20/M input, $1.20/M output), near-frontier performance at roughly 6 cents on the dollar per task. Supports tools and multi-step workflows.

Alibaba's newest flagship in the Qwen family: 2.4T-parameter MoE (95B active), focused on coding and agentic cowork long-horizon tasks. Open weights Qwen3.8-2.4T-A95B (BF16/FP8) released on Hugging Face Aug 13, 2026 — the largest open-weight model ever; the open version is text-only with mandatory thinking and 262K native context. PaperBench 93.0, the highest published score. API $2/$6 per M tokens.

OpenAI's professional-grade model offering advanced reasoning for enterprise workloads

GPT-5.5 low-latency version, fast responses ideal for chat scenarios

3B dense reasoning model. AIME26: 94.3, based on Qwen2.5. Uses Spectrum-to-Signal post-training. No tool calling support, focused on math and code reasoning.

DeepSeek V4 Flash Vision experimental, multimodal input support, extremely low pricing

xAI's latest coding and agentic model based on 1.5T V9 foundation, 500K context, $2/$6/M tokens, coding performance competitive with GPT-5.5

ByteDance Seed 2.1 Turbo model, 262K context, fast speed and low pricing

Cursor AI IDE built-in Agent coding mode, 256K context

Anthropic Claude Sonnet series, mid-to-high-end reasoning model

MiniMax's latest flagship model with competitive performance across reasoning tasks

Qwen3.8 series open model (released Aug 14, 2026, Apache 2.0): 27B dense multimodal model with native image/video understanding, 262K native context (extendable to 1M via YaRN), Gated DeltaNet hybrid architecture, thinking mode on by default and toggleable per request. SWE-bench Pro 61.7, QwenSWEBench 79.0, DeepSWE 1.1 42.2, LiveCodeBench v6 90.3, OSWorld 84.3 - beats Claude Opus 4.6 Max on multiple coding and agent benchmarks. Runs locally on consumer hardware (~48 tps q4km on a 4090).

Meta Superintelligence Labs' open 30B on-device agentic model (Apache 2.0), distilled from Muse Spark, quantized to under 20GB for single-GPU local runs, with multimodal input and tool calling.

OpenAI GPT-4o multimodal model with 128K context

Alibaba's flagship Qwen model with improved instruction following and tool use

Zhipu AI GLM series previous-gen flagship, 200K context

Cursor Composer 2 AI coding assistant

Google DeepMind lightweight Gemini model with 1000K context

Alibaba Qwen3.5 397B parameter large model

OpenAI's compact model in the GPT-5.4 series, optimized for efficiency

Open-weights MoE from Thinking Machines Lab, 276B total/12B active, comparable performance to Inkling at a quarter of the size. Native audio+image reasoning, variable thinking effort, up to 1M context. Highest-scoring open-weight model on ARC Prize.

xAI previous-gen Grok model with 1000K context

ByteDance Seed 2.0 Code model optimized for frontend development, multilingual coding, and agentic coding

DeepSeek's capable mid-range model offering strong performance at lower cost

OpenAI real-time voice conversation model, supports interruption and continuation, simultaneous voice chat and reasoning

Google's lightweight Gemini model optimized for speed and efficiency on edge devices

StepFun's latest multimodal MoE model with 196B parameter language backbone and vision encoder for native image/video understanding

Poolside Laguna XS 2.1 is a 33B-A3B MoE coding agent model optimized for local deployment. Builds on XS.2 with improved SWE-bench Multilingual (63.1%) and stronger terminal-style task performance. Supports 256K context, runs locally in vLLM/SGLang.

Cursor Composer 1.5 earlier version

DeepSeek R1 reasoning-enhanced model, 128K context, MIT open-source

NVIDIA Nemotron 3 Super open-source model

Mistral AI's flagship large language model with strong reasoning and multilingual capabilities

Cohere North Mini Code is Cohere's first agentic coding model and the debut of its North family. A sparse MoE with 30B total / 3B active parameters, optimized for end-to-end coding tasks. 256K context window, text input only.

ByteDance Doubao Seed Code coding model

xAI's coding-focused model trained for agentic software engineering workflows, supports text+image input

Meta Llama 4 Maverick open-source model with 1000K context

IBM's open-source 8B dense enterprise model, matching 32B MoE performance, 131K context, designed for enterprise tasks