AI Model Benchmark Directory

LMSpeed is an AI model directory for comparing API price, output speed, first-token latency, provider coverage, and benchmark data. Use it to narrow your model and provider choice, then test the endpoint that fits your workload.

Price, speed, latency, and availability can change. Treat this table as a current signal and verify your own endpoint before you deploy.

Newest Most tested Recently tested A–Z

Model	Context	Input	Output	Providers	Agents	Coding	Reasoning	Knowledge	Math	Multilingual	Multimodal	Instruction following	Throughput	Latency	Release date
DeepSeek V4 Pro 0813DeepSeek	Context1.0M	Input—	Output—	Providers—	58.6±7.3	56.6±6.1	60.9±8.3	58.4±18.0P	58.9±12.2E	—	—	54.9±16.0P	Throughput —	Latency —	Release date2026-08-12	58.6 56.6 60.9 58.4 58.9 — — 54.9
DeepSeek V4 Flash 0731DeepSeek	Context1.3M	Input—	Output—	Providers	51.9±8.4	52.6±6.3	59.4±8.3	47.5±18.0P	57.9±12.2E	—	—	—	Throughput 74 t/s	Latency 2.70s	Release date2026-07-31	51.9 52.6 59.4 47.5 57.9 — — —
DeepSeek V3DeepSeek	Context163.8K	Input—	Output—	Providers +15	—	—	—	—	—	—	—	—	Throughput —	Latency —	Release date2024-12-26	— — — — — — — —
R1 Distill Llama 70BDeepSeek	Context8.2K	Input$0.700/M	Output$1.10/M	Providers	—	42.6±13.9P	45.3±11.1E	—	55±11.8E	—	—	—	Throughput —	Latency —	Release date2025-01-23	— 42.6 45.3 — 55 — — —
R1 0528DeepSeek	Context163.8K	Input—	Output—	Providers	—	—	—	—	—	—	—	—	Throughput —	Latency —	Release date2025-05-28	— — — — — — — —
DeepSeek V3.2 ExpDeepSeek	Context163.8K	Input$0.280/M	Output$0.420/M	Providers	—	53.3±13.9P	53.9±11.1E	—	52.7±16.1P	—	—	—	Throughput —	Latency —	Release date2025-09-29	— 53.3 53.9 — 52.7 — — —
DeepSeek V4 ProDeepSeek	Context1.0M	Input$0.435/M	Output$0.870/M	Providers +211	50.8±10.5E	47.5±6.6	54.8±9.0	52.6±18.0P	31.6±12.2E	—	—	—	Throughput 59 t/s	Latency 9.81s	Release date2026-04-24	50.8 47.5 54.8 52.6 31.6 — — —
DeepSeek V4 FlashDeepSeek	Context1.0M	Input$0.130/M	Output$0.280/M	Providers +221	45.9±6.3	45.6±6.6	53.6±9.0	41.4±18.0P	32.6±12.2E	—	—	—	Throughput 83 t/s	Latency 6.80s	Release date2026-04-24	45.9 45.6 53.6 41.4 32.6 — — —
DeepSeek V4DeepSeek	Context—	Input—	Output—	Providers +9	—	—	—	—	—	—	—	—	Throughput 47 t/s	Latency 2.49s	Release date—	— — — — — — — —
DeepSeek R1 Distill QwenDeepSeek	Context128K	Input—	Output—	Providers +51	—	40.3±13.9P	39.1±8.9	54.3±16.0P	55.7±11.8E	—	—	37.5±16.0P	Throughput 50 t/s	Latency 19.88s	Release date2025-01-29	— 40.3 39.1 54.3 55.7 — — 37.5
DeepSeek R1 Distill Qwen 1.5BDeepSeek	Context—	Input—	Output—	Providers +21	—	25.6±13.9P	29.7±11.1E	—	45.5±11.8E	—	—	—	Throughput 126 t/s	Latency 4.28s	Release date—	— 25.6 29.7 — 45.5 — — —
DeepSeek V2.5DeepSeek	Context—	Input—	Output—	Providers +4	—	—	41±11.8E	—	45.7±16.0P	—	—	—	Throughput 16 t/s	Latency 0.96s	Release date—	— — 41 — 45.7 — — —
DeepSeek V3.2DeepSeek	Context163.8K	Input$0.280/M	Output$0.420/M	Providers +181	39±6.9	51.1±8.6	50.9±8.7	39.6±16.0P	49±16.1P	—	—	46±16.0P	Throughput 46 t/s	Latency 5.33s	Release date2025-12-01	39 51.1 50.9 39.6 49 — — 46
DeepSeek V3.1DeepSeek	Context—	Input$0.570/M	Output$1.68/M	Providers +119	40.9±16.0P	53.8±13.9P	48.7±8.7	40.7±16.0P	53±16.1P	—	—	42.7±16.0P	Throughput 103 t/s	Latency 1.78s	Release date—	40.9 53.8 48.7 40.7 53 — — 42.7
DeepSeek V3 TurboDeepSeek	Context—	Input—	Output—	Providers +2	—	—	—	—	—	—	—	—	Throughput 28 t/s	Latency 1.31s	Release date—	— — — — — — — —
DeepSeek R1 TurboDeepSeek	Context—	Input—	Output—	Providers +3	—	—	—	—	—	—	—	—	Throughput —	Latency —	Release date—	— — — — — — — —
DeepSeek Chat V3.1DeepSeek	Context163.8K	Input—	Output—	Providers +9	—	—	—	—	—	—	—	—	Throughput 23 t/s	Latency 2.02s	Release date2025-08-21	— — — — — — — —
DeepSeek Coder v2 InstructDeepSeek	Context—	Input—	Output—	Providers +2	—	—	—	—	—	—	—	—	Throughput —	Latency —	Release date—	— — — — — — — —
DeepSeek Prover v2DeepSeek	Context—	Input—	Output—	Providers +8	—	—	—	—	—	—	—	—	Throughput —	Latency —	Release date—	— — — — — — — —
DeepSeek Coder InstructDeepSeek	Context—	Input—	Output—	Providers +4	—	—	—	—	—	—	—	—	Throughput —	Latency —	Release date—	— — — — — — — —
Sonar Reasoning ProPerplexity	Context128K	Input—	Output—	Providers +7	—	—	—	—	57.4±11.8E	—	—	—	Throughput —	Latency —	Release date2025-03-07	— — — — 57.4 — — —
DeepSeek OCRDeepSeek	Context—	Input—	Output—	Providers +41	—	—	—	—	—	—	—	—	Throughput 284 t/s	Latency 0.74s	Release date—	— — — — — — — —
DeepSeek V3.1 TerminusDeepSeek	Context163.8K	Input$0.270/M	Output$1.00/M	Providers +90	—	54.8±13.9P	54.1±11.1E	—	53±16.1P	—	—	—	Throughput 58 t/s	Latency 1.50s	Release date2025-09-22	— 54.8 54.1 — 53 — — —
DeepSeek ReasonerDeepSeek	Context—	Input—	Output—	Providers +85	—	—	—	—	—	—	—	—	Throughput 28 t/s	Latency 6.40s	Release date—	— — — — — — — —
DeepSeek R1DeepSeek	Context64K	Input$1.35/M	Output$3.00/M	Providers +149	41.2±16.0P	49.6±13.9P	50.2±8.7	44.6±16.0P	57±11.8E	—	—	43.3±16.0P	Throughput 52 t/s	Latency 10.36s	Release date2025-05-28	41.2 49.6 50.2 44.6 57 — — 43.3
DeepSeek VL2	Context—	Input—	Output—	Providers +8	—	—	—	—	—	—	—	—	Throughput 126 t/s	Latency 0.75s	Release date—	— — — — — — — —
DeepSeek Coder	Context—	Input—	Output—	Providers +8	—	—	—	—	—	—	—	—	Throughput —	Latency —	Release date—	— — — — — — — —
DeepSeek V3	Context—	Input$0.360/M	Output$0.890/M	Providers +151	35.2±11.8E	37.3±11.2E	44.3±8.7	40.9±16.0P	49.5±9.3E	—	—	39.7±11.9E	Throughput 49 t/s	Latency 3.39s	Release date—	35.2 37.3 44.3 40.9 49.5 — — 39.7

How to read category scores

Agents, Coding, Reasoning, and the other capability columns are 0–100 observed-capability estimates relative to the eligible model population in a dated Category Score V3 run.

They are not success rates, IQ scores, or an average across all eight categories. Read them with the 80% interval, measured dimensions, benchmark families, and evidence shown on each model page.

Read the complete scoring methodology

What this directory shows

Each row brings together the information you need to compare an LLM API. Some fields are blank when LMSpeed has no current data for that model or provider.

API price: Compare input and output price per million tokens when it is available.
Speed and latency: Use throughput and first-token latency to compare response behavior.
Provider coverage: Open a model to review its listed providers and their current details.
Capability data: Use the capability columns when current model scores are available.

How to use the model directory

Start with the decision that matters most for your workload, then compare the current rows before you test an endpoint.

Find a model.Search by model name, slug, or description.
Sort the key metric.Sort by price, throughput, latency, provider count, or capability data.
Compare providers.Open a model page to review the provider options and current data.
Test your endpoint.Run a speed test before you use an endpoint in production.

Model pricing leaderboard Model speed leaderboard Latency leaderboard Provider directory LLM API speed test

Frequently Asked Questions

How does LMSpeed benchmark AI models?

LMSpeed runs standardized five-round API speed tests on each model, measuring output throughput (tokens per second), first-token latency, and total response time across multiple providers.

Which AI model has the lowest API latency?

Latency varies by provider and model. Use the LMSpeed model directory to sort by latency and find the model with the fastest first-token response time. Check the latency leaderboard for monthly rankings.

How to compare LLM API pricing across providers?

LMSpeed lists input and output token prices per million for each model across available providers. Sort by price to find a lower-cost option, or filter by provider to compare rates side by side.

What is an AI model directory?

An AI model directory is a searchable list of models and their comparison data. On LMSpeed, it brings together API price, throughput, first-token latency, provider coverage, and capability data when available.

How do I choose a model and provider?

Start with the metric that matters most for your workload. Then open a model page to compare provider data. Run your own speed test before you use an endpoint in production.

Can model price and speed change?

Yes. Price, availability, throughput, and latency can change by model, provider, and time. Use current page values as a comparison signal and confirm with your own endpoint test.