LLM Leaderboard
Find the right LLM for your workload
Compare leading models across benchmark performance, speed, and cost to understand the trade-offs between them.
Reasoning (GPQA Diamond)
94.8%
Gemini 3.7 Flash (High)
94.60%
GPT-5.4 Pro (xhigh)
94.40%
Gemini 3.1 Pro Preview (high)
94.10%
Gemini 3.6 Flash (high)
94.00%
Grok 4.6 (high)
Coding (SWE-bench Verified)
83.50%
Claude Opus 4.7 (max)
80.60%
GPT-5.5 (xhigh)
79.30%
Gemini 3.5 Flash (high)
78.70%
Claude Opus 4.6 (no thinking)
78.70%
GLM-5.2 (max)
Tool Use (BFCLV4)
77.47%
Claude Opus 4.5 (FC)
73.24%
Claude Sonnet 4.5 (FC)
72.51%
Gemini 3 Pro Preview (Prompt)
72.38%
GLM-4.6 (FC thinking)
69.57%
Grok-4.1 Fast Reasoning (FC)
Fastest Models
1522
Celeris-1
909
Mercury 2
399
Step 3.7 Flash
370
Ling 3.0 Flash
368
Gemini 3.5 Flash-Lite
tokens/second
Lowest Latency
(TTFT)
0.4
Command A+
0.572
North Mini Code
0.6
Celeris-1
0.65
Ministral 3 3B
0.74
Gemini 3.7 Flash
Cheapest Models
Input Cost
Output Cost
$0.16
$0.12
$0.08
$0.04
$0
Gemma 4 E4B
Sarvam 30B
Nova Micro
Qwen3.5 4B
Nemotron Nano 9B V2
FAQ
Frequently asked questions
What is the LLM Leaderboard, and how does it work?
The LLM Leaderboard compares leading models across capability, speed, latency, context window, and cost. It brings benchmark and performance data into one place so you can compare models without relying on a single score.
What benchmarks are used to evaluate models on the LLM Leaderboard?
We use GPQA Diamond for reasoning, SWE-bench Verified for coding, and BFCL V4 for tool use. Each benchmark measures a different capability, so the scores are shown separately rather than combined into one overall ranking.
How does the LLM Leaderboard rank models based on speed, cost, and latency?
Speed is measured using median output throughput in tokens per second, while latency is measured by time to first token or first chunk. Cost comparisons use current input and output token pricing so you can see how model performance affects operating cost.
What details are included in the model comparison table?
The comparison table includes each model’s:
Provider
Release date
Context window
Input and output cost
Available benchmark scores (GPQA, SWE-bench Verified, BFCL V4)
Where a model hasn’t been evaluated by a benchmark source, the score is left blank rather than replaced with an incomparable result.
How can I use the LLM Leaderboard to choose the best model for my use case?
Start with the capability that matters most for your workload like reasoning, coding, or tool use. Then compare context limits, latency, throughput, and cost to find the model that gives you the best overall trade-off for your application.
