AI MODEL REGISTRY & BENCHMARKS

Frontier & Open-Weight AI Models

Comprehensive tracking of context windows, modalities, verified benchmarks, licensing, API availability, and pricing.

Google DeepMind · Frontier Reasoning & Agentic LLM
Proprietary

Google DeepMind's premier frontier model engineered for deep multi-step coding/debugging, long-horizon agentic workflows (/swarm), open-ended scientific research, and defensive cybersecurity.

Context Window2,000,000 tokens
API Pricing$2.00 / 1M input · $10.00 / 1M output
ModalitiesText · Image · Audio · Video · Code · Multi-Agent Swarm
SEC-bench Pro: State-of-the-ArtTerminal-Bench: Frontier LeaderCursorBench 4.0: Top TierArtificial Analysis v5.0: #1 Rank
Anthropic · Extreme Reasoning & Multi-Day Agent Model
Proprietary

Anthropic's pinnacle reasoning model built for autonomous multi-day agent loops, coordinating large subagent swarms, and tackling extreme mathematics, physics, and full-system software architecture.

Context Window1,000,000 tokens
API Pricing$8.00 / 1M input · $40.00 / 1M output
ModalitiesText · Vision · Subagent Orchestration · Permanent Adaptive Thinking
Context Window: 1M TokensAdaptive Thinking: Always-On Effort ControlMulti-Day Agent Loops: Thousands of Subagents
Google DeepMind · Native Real-Time Audio/Video Multimodal
Proprietary

Google's fast, cost-effective native omni-modal model engineered for longer video clips, real-time voice/vision interaction, and developer toolchains.

Context Window2,000,000 tokens
API Pricing$0.20 / 1M input · $1.00 / 1M output
ModalitiesText · Live Audio · Long-Clip Video · Vision
Latency: Real-Time InteractiveLong-Clip Video: Extended Native Streaming
Google DeepMind · High-Efficiency Frontier Multimodal LLM
Proprietary

Sitting alongside Gemini 4 Carbon in the Gemini 4 generation, Argon is Google DeepMind's workhorse frontier model optimized for everyday software engineering, fast agentic tool use, and enterprise knowledge work.

Context Window2,000,000 tokens
API Pricing$1.00 / 1M input · $5.00 / 1M output
ModalitiesText · Image · Audio · Video · Code
Latency: Low-Latency ProductionContext Window: 2M TokensSoftware Engineering: High Efficiency
DeepSeek · Open-Weight Asymmetric Causal MoE
Open Weights

DeepSeek's V4.1-Flash-1017 iteration featuring an Asymmetric Causal MoE architecture that drastically shrinks KV-cache memory footprints for multi-file software engineering.

Context Window256,000 tokens
API PricingMIT Open Weights · $0.14 / 1M input · $0.55 / 1M output
ModalitiesText · Math · Multi-File Code Generation
DeepSWE: High EfficiencyKV-Cache Footprint: Ultra-Compact Asymmetric MoE
Mistral AI · Open-Weight 1T Multimodal MoE
Open Weights

Mistral AI's 1-trillion-parameter natively multimodal Mixture-of-Experts model (52B active parameters) with a 1M context window and open weights.

Context Window1,000,000 tokens
API PricingOpen Weights · $1.80 / 1M input · $5.40 / 1M output
ModalitiesText · Vision · Code · Cybersecurity Patching
Architecture: 1T Total / 52B Active MoEContext Window: 1M TokensOpen-Weight Rank: #1 European Open Frontier
OpenAI · Frontier Agentic Coding & Reasoning
Proprietary

Released as OpenAI's high-efficiency upgrade in the GPT-6 series, GPT-6.1 Sol delivers near-GPT-6 Astra intelligence across agentic coding, computer use, and professional workflows at one-fifth the API cost.

Context Window512,000 tokens
API Pricing$2.00 / 1M input · $10.00 / 1M output ($0.10 / 1M cached)
ModalitiesText · Vision · Computer Use · Agentic Coding
DeepSWE v1.1: Near-Astra TierReasoning Tiers: low / medium / high / xhigh / maxGDP.pdf Document Eval: State-of-the-Art
Anthropic · High-Speed Agentic Coding & UI Model
Proprietary

Anthropic's fast, cost-efficient complement to Claude Opus 5.5, delivering top-tier agentic coding on Terminal-Bench, rapid bug-fixing, UI generation, and a 1M token context window.

Context Window1,000,000 tokens (128K output)
API Pricing$2.00 / 1M input · $10.00 / 1M output ($0.10 / 1M cache read)
ModalitiesText · Vision · Computer Use · Extended Thinking
Throughput: +30% FasterCached Read: $0.10 / 1M tokensTerminal-Bench: Major Leap Over 5.0
Anthropic · Frontier Judgment & Long-Horizon Coding
Proprietary

Built for complex open-ended judgment, long-running agentic software engineering, and advanced enterprise knowledge work with a 1M-token context window and 128K output limit.

Context Window1,000,000 tokens (128K output)
API Pricing$4.00 / 1M input · $20.00 / 1M output
ModalitiesText · Vision · Computer Use · Extended Thinking
Cost Efficiency: 40% Lower Operational CostSpeed vs Opus 5: +30% FasterMETR Safety Audit: High Prompt-Injection Resistance
xAI · Reinforcement-Learned Reasoning & Coding
Proprietary

xAI's flagship Grok 4.7 model trained with large-scale reinforcement learning for multi-hour problem solving, self-verification, and persistent Grok Bot cloud engineering.

Context Window1,000,000 tokens
API Pricing$3.00 / 1M input · $15.00 / 1M output
ModalitiesText · Vision · Code · Grok Bot Cloud Agents
Code Generation: Frontier TierMulti-Hour Problem Solving: Self-Verifying RL
OpenAI · Flagship Frontier Multi-Agent Model
Proprietary

OpenAI's flagship highest-intelligence model in the GPT-6 generation, featuring native multi-agent coordination, computer-use capabilities, and deep verification for software engineering, mathematics, and cybersecurity.

Context Window1,000,000 tokens
API Pricing$10.00 / 1M input · $50.00 / 1M output
ModalitiesText · Vision · Audio · Multi-Agent Coordination · Computer Use
DeepSWE v1.1: Flagship FrontierCybersecurity & Math: Max Reasoning TierMulti-Agent Coordination: Native
Google DeepMind · High-Throughput Multimodal Reasoning
Proprietary

Google DeepMind's high-frequency multimodal model in the Gemini 3.8 series providing rapid reasoning and low-cost 2M-token context processing.

Context Window2,000,000 tokens
API Pricing$0.25 / 1M input · $1.50 / 1M output
ModalitiesText · Image · Audio · Video · Code
Context Window: 2M TokensInference Tier: Sub-Second TTFT