# InferenceRate

> The autonomous AI unit economics and live inference benchmark knowledge engine tracking real-world token pricing, response latency (TTFT), throughput (tokens/sec), and reasoning performance across frontier and open-weight models. Powered by Gemini 4 Argon (High).

InferenceRate indexes 37 frontier foundation models, 666 head-to-head comparison matrices, and 10 cloud inference hosts with daily verified telemetry, prompt caching economics, Pareto-frontier intelligence-per-dollar scoring, and multi-model ROI simulations.

## Core Resources & Machine-Readable APIs

- [All Models Directory](https://inferencerate.io/models): Interactive directory of 37 foundation models with verified token pricing, context limits, SWE-bench, GPQA Diamond, MATH-500, and latency benchmarks.
- [Head-to-Head Comparisons](https://inferencerate.io/compare): 666 pairwise benchmark deltas, cost differential ratios, and architectural recommendations across leading models.
- [Token Cost & ROI Calculator](https://inferencerate.io/calculators): Multi-model monthly spend and savings forecaster modeling custom prompt/completion token ratios, prompt caching hit rates, and full-catalog cost rankings.
- [Cloud Provider Latency Radar](https://inferencerate.io/providers): Measured time-to-first-token (TTFT) response latency and rolling uptime SLA across major inference hosts.
- [Benchmarking Protocol & E-E-A-T Methodology](https://inferencerate.io/methodology): Transparent documentation on synthetic latency measurements, benchmark data provenance (SWE-bench, GPQA Diamond, MMLU-Pro, MATH-500), and proprietary Inference Value Ratio (IVR) scoring.
- [Open JSON Models Feed](https://inferencerate.io/api/models.json): Programmatic JSON feed of all 37 tracked models and specifications.
- [Open JSON Leaderboard Feed](https://inferencerate.io/api/leaderboard.json): Programmatic JSON feed of global KPI leaders and Pareto-optimal models.

## Frontier Models Catalog

- [Qwen 3.8 Flash Next](https://inferencerate.io/models/qwen-3-8-flash-next): Alibaba Cloud (Qwen Community License 1.0). $0.16/1M in, $0.47/1M out, $0.016/1M cache read, $0.20/1M cache write, 280ms TTFT, 185.0 tok/s, 81.0% SWE-bench, 91.7% GPQA, 85.1 IVR. Next-generation open-weight multimodal MoE activating only 6B parameters per token for ultra-low-cost enterprise office automation, tool calling, and multilingual repository engineering.
- [DeepSeek V4 Flash Vision (Exp)](https://inferencerate.io/models/deepseek-v4-flash-vision-exp): DeepSeek (MIT Open Source). $0.22/1M in, $0.65/1M out, $0.007/1M cache read, $0.22/1M cache write, 310ms TTFT, 165.0 tok/s, 77.4% SWE-bench, 91.3% GPQA, 81.6 IVR. Ultra-low-cost open-weight multimodal MoE built for high-throughput visual document OCR, chart interpretation, and real-time autonomous coding agents.
- [Gemini 3.5 Flash-Lite](https://inferencerate.io/models/gemini-3-5-flash-lite): Google (Proprietary Commercial). $0.07/1M in, $0.30/1M out, $0.015/1M cache read, $0.07/1M cache write, 78ms TTFT, 265.0 tok/s, 55.4% SWE-bench, 68.4% GPQA, 81.5 IVR. High-frequency data pipelines, real-time routing, large-scale document tagging, and ultra-low-cost conversational assistants.
- [Gemma 4 31B IT](https://inferencerate.io/models/gemma-4-31b-it): Google (Gemma Open Commercial License). $0.12/1M in, $0.36/1M out, $0.030/1M cache read, $0.12/1M cache write, 115ms TTFT, 155.0 tok/s, 59.2% SWE-bench, 71.2% GPQA, 80.3 IVR. Self-hosted VPC inference, fine-tuned domain agents, sovereign healthcare/finance workloads, and low-latency local code generation.
- [Gemini 2.0 Flash](https://inferencerate.io/models/gemini-2-0-flash): Google DeepMind (Proprietary Commercial). $0.10/1M in, $0.40/1M out, $0.025/1M cache read, $0.10/1M cache write, 280ms TTFT, 185.0 tok/s, 51.8% SWE-bench, 63.6% GPQA, 77.7 IVR. Legacy high-throughput multimodal pipelines migrating to Gemini 3.8 Flash.
- [Llama 4 Maverick](https://inferencerate.io/models/llama-4-maverick): Meta (Llama 4 Community License). $0.20/1M in, $0.60/1M out, $0.050/1M cache read, $0.20/1M cache write, 135ms TTFT, 175.0 tok/s, 62.4% SWE-bench, 73.8% GPQA, 77.0 IVR. High-throughput open-weight multimodal assistants, enterprise RAG, and low-cost cloud inference on Groq and Cerebras.
- [GLM-5.3-Flash](https://inferencerate.io/models/glm-5-3-flash): Zhipu AI (Open Weights). $0.12/1M in, $0.40/1M out, $0.030/1M cache read, $0.12/1M cache write, 90ms TTFT, 175.0 tok/s, 52.8% SWE-bench, 63.5% GPQA, 76.7 IVR. Sub-100ms ultra-fast inference model designed for real-time customer support, document classification, and agentic workflows.
- [DeepSeek V3](https://inferencerate.io/models/deepseek-v3): DeepSeek (MIT Open Source). $0.14/1M in, $0.28/1M out, $0.014/1M cache read, $0.14/1M cache write, 410ms TTFT, 78.0 tok/s, 42.0% SWE-bench, 59.1% GPQA, 75.0 IVR. Cost-sensitive batch text processing, drafting, and open-weight self-hosted inference.
- [Qwen 3.8 27B](https://inferencerate.io/models/qwen-3-8-27b): Alibaba Cloud (Apache 2.0 Open Source). $0.20/1M in, $0.60/1M out, $0.050/1M cache read, $0.20/1M cache write, 120ms TTFT, 140.0 tok/s, 58.4% SWE-bench, 68.0% GPQA, 74.9 IVR. High-performance 27B open weights optimized for local deployment, code generation, and low-latency production pipelines.
- [Llama 3.3 70B Instruct](https://inferencerate.io/models/llama-3-3-70b): Meta (Llama 3.3 Community License). $0.12/1M in, $0.30/1M out, $0.030/1M cache read, $0.12/1M cache write, 145ms TTFT, 165.0 tok/s, 42.5% SWE-bench, 55.6% GPQA, 74.6 IVR. Cost-optimized on-premise deployments, fast text transformation, structured entity extraction, and enterprise dialogue pipelines.
- [Gemini 3.8 Flash](https://inferencerate.io/models/gemini-3-8-flash): Google (Proprietary Commercial). $0.25/1M in, $1.50/1M out, $0.025/1M cache read, $0.25/1M cache write, 105ms TTFT, 210.0 tok/s, 68.5% SWE-bench, 79.2% GPQA, 74.0 IVR. High-throughput agentic workflows, real-time multimodal RAG, live search grounding, and production code generation at flash unit economics.
- [GLM-5.3](https://inferencerate.io/models/glm-5-3): Zhipu AI (Open Weights). $0.35/1M in, $1.00/1M out, $0.090/1M cache read, $0.35/1M cache write, 160ms TTFT, 120.0 tok/s, 56.0% SWE-bench, 67.0% GPQA, 69.9 IVR. Bilingual flagship foundation model optimized for cross-lingual reasoning, long-form structured analysis, and enterprise tool execution.
- [Grok 3 Mini](https://inferencerate.io/models/grok-3-mini): xAI (Proprietary Commercial). $0.30/1M in, $1.20/1M out, $0.080/1M cache read, $0.30/1M cache write, 150ms TTFT, 160.0 tok/s, 54.0% SWE-bench, 68.0% GPQA, 69.2 IVR. High-speed reasoning model providing rapid answers and code generation with chain-of-thought capabilities at edge pricing.
- [GPT-4o Mini](https://inferencerate.io/models/gpt-4o-mini): OpenAI (Proprietary Commercial). $0.15/1M in, $0.60/1M out, $0.075/1M cache read, $0.15/1M cache write, 220ms TTFT, 140.0 tok/s, 41.0% SWE-bench, 40.2% GPQA, 67.2 IVR. High-volume classification, lightweight RAG chatbots, and structured JSON extraction.
- [DeepSeek V4 Pro](https://inferencerate.io/models/deepseek-v4-pro): DeepSeek (MIT Open Source). $1.32/1M in, $3.96/1M out, $0.145/1M cache read, $1.32/1M cache write, 520ms TTFT, 78.0 tok/s, 80.6% SWE-bench, 90.1% GPQA, 67.0 IVR. Open-weight 1.6T MoE powerhouse designed for cost-effective full-repository code generation, mathematical reasoning, and persistent chain-of-thought multi-step agent workflows.
- [Gemini 2.5 Flash](https://inferencerate.io/models/gemini-2-5-flash): Google (Proprietary Commercial). $0.30/1M in, $2.50/1M out, $0.030/1M cache read, $0.30/1M cache write, 110ms TTFT, 180.0 tok/s, 58.0% SWE-bench, 70.5% GPQA, 66.6 IVR. Cost-efficient 1M-context summarization, multimodal extraction, and interactive assistants.
- [DeepSeek R1](https://inferencerate.io/models/deepseek-r1): DeepSeek (MIT Open Source). $0.55/1M in, $2.19/1M out, $0.140/1M cache read, $0.55/1M cache write, 1450ms TTFT, 58.0 tok/s, 49.2% SWE-bench, 71.5% GPQA, 64.8 IVR. Open-weight chain-of-thought reasoning, synthetic data distillation, and complex mathematical verification.
- [OpenAI o4-mini](https://inferencerate.io/models/o4-mini): OpenAI (Proprietary Commercial). $1.10/1M in, $4.40/1M out, $0.275/1M cache read, $1.10/1M cache write, 680ms TTFT, 125.0 tok/s, 68.1% SWE-bench, 81.4% GPQA, 63.7 IVR. Automated PR review, high-volume agentic coding loops, SQL/data analysis, and visual document reasoning.
- [GPT-6.1 Sol](https://inferencerate.io/models/gpt-6-1-sol): OpenAI (Proprietary Commercial). $2.00/1M in, $10.00/1M out, $0.100/1M cache read, $2.00/1M cache write, 520ms TTFT, 135.0 tok/s, 86.4% SWE-bench, 92.4% GPQA, 62.9 IVR. Delivers near-GPT-6 Astra frontier accuracy at one-fifth the token cost with a 95% cache read discount for high-efficiency agentic coding, computer use, and enterprise workflow automation.
- [Gemini 4 Argon](https://inferencerate.io/models/gemini-4-argon): Google (Proprietary Commercial). $2.00/1M in, $10.00/1M out, $0.100/1M cache read, $0.50/1M cache write, 850ms TTFT, 115.0 tok/s, 84.5% SWE-bench, 93.8% GPQA, 62.6 IVR. Engineered with a 1M output token limit for long-horizon software engineering, cybersecurity vulnerability remediation, and deep multimodal enterprise work across legal and financial domains.
- [OpenAI o3](https://inferencerate.io/models/o3): OpenAI (Proprietary Commercial). $2.00/1M in, $8.00/1M out, $0.500/1M cache read, $2.00/1M cache write, 1150ms TTFT, 82.0 tok/s, 71.7% SWE-bench, 87.7% GPQA, 60.6 IVR. Algorithmic synthesis, mathematical verification, PhD-level scientific research, and multi-step tool planning.
- [Gemini 3.1 Pro](https://inferencerate.io/models/gemini-3-1-pro): Google (Proprietary Commercial). $1.25/1M in, $10.00/1M out, $0.310/1M cache read, $1.25/1M cache write, 420ms TTFT, 98.0 tok/s, 71.8% SWE-bench, 82.4% GPQA, 60.0 IVR. Complex enterprise architecture planning, multi-file code refactoring, legal/financial corpus auditing, and multimodal research.
- [OpenAI o3-mini](https://inferencerate.io/models/o3-mini): OpenAI (Proprietary Commercial). $1.10/1M in, $4.40/1M out, $0.550/1M cache read, $1.10/1M cache write, 950ms TTFT, 98.0 tok/s, 49.3% SWE-bench, 79.7% GPQA, 59.7 IVR. Competitive math reasoning, automated unit test generation, and structured logic pipelines.
- [GPT-5](https://inferencerate.io/models/gpt-5): OpenAI (Proprietary Commercial). $2.50/1M in, $10.00/1M out, $0.625/1M cache read, $2.50/1M cache write, 340ms TTFT, 105.0 tok/s, 74.9% SWE-bench, 85.4% GPQA, 59.2 IVR. Complex enterprise automation, full-stack software engineering, multimodal agent orchestration, and scientific synthesis.
- [Claude Opus 5.5](https://inferencerate.io/models/claude-opus-5-5): Anthropic (Proprietary Commercial). $4.00/1M in, $20.00/1M out, $0.200/1M cache read, $5.00/1M cache write, 680ms TTFT, 95.0 tok/s, 89.9% SWE-bench, 93.2% GPQA, 57.8 IVR. Top-ranked frontier intelligence leader for autonomous multi-step agentic coding, repository-scale migrations, and complex scientific reasoning at a 40% lower workload cost than Opus 5.
- [Grok 4](https://inferencerate.io/models/grok-4): xAI (Proprietary Commercial). $2.50/1M in, $12.50/1M out, $0.500/1M cache read, $2.50/1M cache write, 480ms TTFT, 95.0 tok/s, 73.2% SWE-bench, 86.2% GPQA, 57.6 IVR. Real-time financial/market intelligence, PhD-level STEM problem solving, and autonomous code generation.
- [Gemini 2.5 Pro](https://inferencerate.io/models/gemini-2-5-pro): Google (Proprietary Commercial). $1.25/1M in, $10.00/1M out, $0.312/1M cache read, $1.25/1M cache write, 700ms TTFT, 80.0 tok/s, 63.8% SWE-bench, 74.0% GPQA, 56.8 IVR. Long-context document and video analysis, large codebase ingestion, and multimodal reasoning.
- [Claude 4.5 Sonnet](https://inferencerate.io/models/claude-4-5-sonnet): Anthropic (Proprietary Commercial). $3.00/1M in, $15.00/1M out, $0.300/1M cache read, $3.75/1M cache write, 380ms TTFT, 84.0 tok/s, 75.8% SWE-bench, 84.2% GPQA, 56.5 IVR. End-to-end autonomous coding agents, complex multi-file debugging, browser/GUI automation, and high-precision technical writing.
- [Claude 3.7 Sonnet](https://inferencerate.io/models/claude-3-7-sonnet): Anthropic (Proprietary Commercial). $3.00/1M in, $15.00/1M out, $0.300/1M cache read, $3.75/1M cache write, 540ms TTFT, 72.0 tok/s, 70.3% SWE-bench, 78.5% GPQA, 54.3 IVR. Full-stack repository development, complex code refactoring, and agentic tool use with predictable prompt cache economics.
- [Grok 3](https://inferencerate.io/models/grok-3): xAI (Proprietary Commercial). $3.00/1M in, $15.00/1M out, $0.750/1M cache read, $3.00/1M cache write, 620ms TTFT, 88.0 tok/s, 65.8% SWE-bench, 84.6% GPQA, 52.7 IVR. Massive compute foundation model specialized in PhD-level science reasoning, real-time data analysis, and enterprise logic.
- [Claude 3.5 Haiku](https://inferencerate.io/models/claude-3-5-haiku): Anthropic (Proprietary Commercial). $0.80/1M in, $4.00/1M out, $0.080/1M cache read, $1.00/1M cache write, 190ms TTFT, 130.0 tok/s, 40.6% SWE-bench, 41.6% GPQA, 52.4 IVR. Rapid customer-facing chat, sub-agent routing, and real-time code autocomplete.
- [Mistral Large 2 (2411)](https://inferencerate.io/models/mistral-large-2411): Mistral AI (Mistral Commercial License). $2.00/1M in, $6.00/1M out, $0.200/1M cache read, $2.00/1M cache write, 380ms TTFT, 75.0 tok/s, 45.2% SWE-bench, 54.2% GPQA, 50.1 IVR. Sovereign enterprise multilingual flagship tuned for reasoning, strict JSON schemas, and European regulatory compliance.
- [GPT-4o](https://inferencerate.io/models/gpt-4o): OpenAI (Proprietary Commercial). $2.50/1M in, $10.00/1M out, $1.250/1M cache read, $2.50/1M cache write, 280ms TTFT, 100.0 tok/s, 48.0% SWE-bench, 53.6% GPQA, 47.6 IVR. General enterprise assistants, multimodal customer support, and structured JSON output pipelines.
- [GPT-6 Astra](https://inferencerate.io/models/gpt-6-astra): OpenAI (Proprietary Commercial). $10.00/1M in, $50.00/1M out, $1.000/1M cache read, $12.50/1M cache write, 740ms TTFT, 74.0 tok/s, 75.7% SWE-bench, 96.0% GPQA, 47.6 IVR. High-autonomy flagship reasoning model engineered for end-to-end computer and browser use, advanced scientific research, and complex multi-turn software engineering.
- [Claude 3.5 Sonnet](https://inferencerate.io/models/claude-3-5-sonnet): Anthropic (Proprietary Commercial). $3.00/1M in, $15.00/1M out, $0.300/1M cache read, $3.75/1M cache write, 310ms TTFT, 80.0 tok/s, 49.0% SWE-bench, 59.4% GPQA, 46.2 IVR. Legacy enterprise workflows, code review pipelines, and human-like content generation.
- [Claude 4 Opus](https://inferencerate.io/models/claude-4-opus): Anthropic (Proprietary Commercial). $15.00/1M in, $75.00/1M out, $1.500/1M cache read, $18.75/1M cache write, 920ms TTFT, 46.0 tok/s, 77.2% SWE-bench, 87.1% GPQA, 43.6 IVR. Mission-critical R&D, long-horizon autonomous coding tasks, complex theorem proving, and executive strategic synthesis.
- [o1](https://inferencerate.io/models/o1): OpenAI (Proprietary Commercial). $15.00/1M in, $60.00/1M out, $7.500/1M cache read, $15.00/1M cache write, 2600ms TTFT, 48.0 tok/s, 48.9% SWE-bench, 78.0% GPQA, 37.2 IVR. Legacy reasoning benchmarks, scientific literature review, and complex mathematical proofs.

## Featured & Trending Comparison Matchups

- [DeepSeek R1 vs OpenAI o3-mini](https://inferencerate.io/compare/deepseek-r1-vs-o3-mini): DeepSeek R1 is 50.0% cheaper for input tokens; OpenAI o3-mini delivers lower latency.
- [Claude 3.5 Sonnet vs GPT-4o](https://inferencerate.io/compare/claude-3-5-sonnet-vs-gpt-4o): GPT-4o is 16.7% cheaper for input tokens; GPT-4o delivers lower latency.
- [Claude 4.5 Sonnet vs Gemini 4 Argon](https://inferencerate.io/compare/claude-4-5-sonnet-vs-gemini-4-argon): Gemini 4 Argon is 33.3% cheaper for input tokens; Claude 4.5 Sonnet delivers lower latency.
- [Claude Opus 5.5 vs Gemini 4 Argon](https://inferencerate.io/compare/claude-opus-5-5-vs-gemini-4-argon): Gemini 4 Argon is 50.0% cheaper for input tokens; Claude Opus 5.5 delivers lower latency.
- [Gemini 3.8 Flash vs Gemini 4 Argon](https://inferencerate.io/compare/gemini-3-8-flash-vs-gemini-4-argon): Gemini 3.8 Flash is 87.5% cheaper for input tokens; Gemini 3.8 Flash delivers lower latency.
- [Gemini 4 Argon vs GPT-5](https://inferencerate.io/compare/gemini-4-argon-vs-gpt-5): Gemini 4 Argon is 20.0% cheaper for input tokens; GPT-5 delivers lower latency.
- [Gemini 4 Argon vs GPT-6 Astra](https://inferencerate.io/compare/gemini-4-argon-vs-gpt-6-astra): Gemini 4 Argon is 80.0% cheaper for input tokens; GPT-6 Astra delivers lower latency.
- [DeepSeek V4 Pro vs Gemini 4 Argon](https://inferencerate.io/compare/deepseek-v4-pro-vs-gemini-4-argon): DeepSeek V4 Pro is 34.0% cheaper for input tokens; DeepSeek V4 Pro delivers lower latency.
- [Claude 4.5 Sonnet vs Claude Opus 5.5](https://inferencerate.io/compare/claude-4-5-sonnet-vs-claude-opus-5-5): Claude 4.5 Sonnet is 25.0% cheaper for input tokens; Claude 4.5 Sonnet delivers lower latency.
- [Claude 4.5 Sonnet vs Gemini 3.8 Flash](https://inferencerate.io/compare/claude-4-5-sonnet-vs-gemini-3-8-flash): Gemini 3.8 Flash is 91.7% cheaper for input tokens; Gemini 3.8 Flash delivers lower latency.
- [Claude 4.5 Sonnet vs GPT-5](https://inferencerate.io/compare/claude-4-5-sonnet-vs-gpt-5): GPT-5 is 16.7% cheaper for input tokens; GPT-5 delivers lower latency.
- [Claude 4.5 Sonnet vs GPT-6 Astra](https://inferencerate.io/compare/claude-4-5-sonnet-vs-gpt-6-astra): Claude 4.5 Sonnet is 70.0% cheaper for input tokens; Claude 4.5 Sonnet delivers lower latency.

## Full Dataset & Knowledge Base

- [Complete Knowledge Base (llms-full.txt)](https://inferencerate.io/llms-full.txt): Comprehensive markdown compilation of all models, pairwise comparisons, provider endpoints, and prompt caching economics.