实时
Apr 7Claude Mythos Preview · Anthropic's most capable model arrives·Mar 31GPT-5.4 Nano launched on OpenAI·Mar 31GPT-5.4 Mini joins the OpenAI lineup·Mar 30Claude Opus 4.5 input price dropped · $5.00 per 1M tokens·Mar 30Mistral Small 4 available via Mistral AI·Mar 29Gemini 2.5 Pro scores 94.1% on MMLU·Mar 29Grok 4.20 Multi-Agent Beta enters agent rankings·Mar 28DeepSeek V3.2 output price dropped · $0.38 per 1M tokens·Mar 287 new MCP servers added in dev-tools category·Mar 27Claude Sonnet 4.6 released by Anthropic·Mar 27Claude Opus 4.6 released by Anthropic·Mar 26OTIS Mock AIME 2024-2025 benchmark added·Mar 26Claude Opus 4.1 pricing increased · $15/$75 per 1M tokens·Mar 25Grok 4.20 Beta launched by xAI·Mar 25Inception added as a tracked provider·Mar 24DeepSeek R1 0528 posted 87.2% on GPQA Diamond·Mar 243 new MCP servers in AI/ML category·Mar 23GPT-4o Audio Preview marked as deprecated·Mar 23Mistral Medium 3.1 input price cut to $0.40 per 1M tokens·Mar 22DeepSeek V3.2 Speciale released·Mar 22WeirdML benchmark now tracked on BenchGecko·Mar 20Nemotron 3 Super (120B) launched by NVIDIA·Mar 20Gemini 2.5 Flash Lite priced at $0.10/$0.40 per 1M tokens·Mar 18Mistral Large 3 2512 released by Mistral AI·Mar 18Grok Code Fast 1 added to agent rankings·Mar 16Claude Sonnet 4.5 scores 91.7% on MMLU·Mar 1612 new MCP servers added across 5 categories·Mar 14GPT-5.4 Pro launched · OpenAI's new flagship·Mar 14GPT-5.4 standard tier released by OpenAI·Mar 12Grok 3 Mini marked as deprecated by xAI·Mar 12Llama 3.3 Nemotron Super 49B pricing dropped·Mar 10Liquid added as a tracked provider·Mar 10MiniMax M2.7 released by MiniMax·Mar 8Grok 4 posted 89.4% on GPQA Diamond·Mar 8LAMBADA benchmark scores now tracked·Mar 5Gemini 2.5 Flash output price reduced to $2.50 per 1M tokens·Mar 5Mercury 2 launched by Inception·Mar 3Qwen3.5-Flash released by Alibaba Qwen·Mar 35 new MCP servers added · finance and auth categories·Apr 7Claude Mythos Preview · Anthropic's most capable model arrives·Mar 31GPT-5.4 Nano launched on OpenAI·Mar 31GPT-5.4 Mini joins the OpenAI lineup·Mar 30Claude Opus 4.5 input price dropped · $5.00 per 1M tokens·Mar 30Mistral Small 4 available via Mistral AI·Mar 29Gemini 2.5 Pro scores 94.1% on MMLU·Mar 29Grok 4.20 Multi-Agent Beta enters agent rankings·Mar 28DeepSeek V3.2 output price dropped · $0.38 per 1M tokens·Mar 287 new MCP servers added in dev-tools category·Mar 27Claude Sonnet 4.6 released by Anthropic·Mar 27Claude Opus 4.6 released by Anthropic·Mar 26OTIS Mock AIME 2024-2025 benchmark added·Mar 26Claude Opus 4.1 pricing increased · $15/$75 per 1M tokens·Mar 25Grok 4.20 Beta launched by xAI·Mar 25Inception added as a tracked provider·Mar 24DeepSeek R1 0528 posted 87.2% on GPQA Diamond·Mar 243 new MCP servers in AI/ML category·Mar 23GPT-4o Audio Preview marked as deprecated·Mar 23Mistral Medium 3.1 input price cut to $0.40 per 1M tokens·Mar 22DeepSeek V3.2 Speciale released·Mar 22WeirdML benchmark now tracked on BenchGecko·Mar 20Nemotron 3 Super (120B) launched by NVIDIA·Mar 20Gemini 2.5 Flash Lite priced at $0.10/$0.40 per 1M tokens·Mar 18Mistral Large 3 2512 released by Mistral AI·Mar 18Grok Code Fast 1 added to agent rankings·Mar 16Claude Sonnet 4.5 scores 91.7% on MMLU·Mar 1612 new MCP servers added across 5 categories·Mar 14GPT-5.4 Pro launched · OpenAI's new flagship·Mar 14GPT-5.4 standard tier released by OpenAI·Mar 12Grok 3 Mini marked as deprecated by xAI·Mar 12Llama 3.3 Nemotron Super 49B pricing dropped·Mar 10Liquid added as a tracked provider·Mar 10MiniMax M2.7 released by MiniMax·Mar 8Grok 4 posted 89.4% on GPQA Diamond·Mar 8LAMBADA benchmark scores now tracked·Mar 5Gemini 2.5 Flash output price reduced to $2.50 per 1M tokens·Mar 5Mercury 2 launched by Inception·Mar 3Qwen3.5-Flash released by Alibaba Qwen·Mar 35 new MCP servers added · finance and auth categories·
AI 经济， 全面追踪。
脉搏19·健康
泡沫278%·波动
GPT-5.5 Pro+4.0
Open Source16.2%
排行榜
GPT-5.5 Pro
#1
99.9
$30.00/M
Qwen3.5 397B A17B
#2
96.3
$0.39/M
DeepSeek V3.2 Speciale
#3
95.2
$0.40/M
GPT-5.4 Pro
#4
93.0
$30.00/M
GPT-5.1-Codex-Max
#5
91.2
$1.25/M
Gemini 3.1 Pro Preview
#6
90.0
$2.00/M
Step 3.5 Flash
#7
89.5
$0.10/M
GPT-5 Chat
#8
89.0
$1.25/M
Qwen3.6 Plus
#9
88.7
$0.33/M
GLM 5.1
#10
87.0
$1.05/M
GPT-5.2-Codex
#11
85.4
$1.75/M
GPT-5.4
#12
83.4
$2.50/M
Claude Opus 4.6 (Fast)
#13
83.3
$30.00/M
柱宽 · 平均基准分数 · 颜色 · 类别
The Pulse
脉搏
19
healthy
7d · +3 分
泡沫指数 · 分项
Valuation Premiumhealthy+2.1
Funding Accelerationhealthy+1.5
Concentration Riskhealthy0
Revenue Qualityhealthy+1.4
Capex Gaphealthy+0.3
最大变动 · Valuation Premium 上升 2.1 分
AI 泡沫指数278%
健康泡沫初现过热泡沫
跨层洞察
最新基准测试变动
全部 →今日 AI
全部 →Apr 7
Claude Mythos Preview · Anthropic's most capable model arrivesMar 31
GPT-5.4 Nano launched on OpenAIMar 31GPT-5.4 Mini joins the OpenAI lineupMar 30Claude Opus 4.5 input price dropped · $5.00 per 1M tokensMar 30
Mistral Small 4 available via Mistral AIMar 29
Gemini 2.5 Pro scores 94.1% on MMLUMar 29
Grok 4.20 Multi-Agent Beta enters agent rankingsMar 28
DeepSeek V3.2 output price dropped · $0.38 per 1M tokensMar 287 new MCP servers added in dev-tools categoryMar 27Claude Sonnet 4.6 released by AnthropicMar 27Claude Opus 4.6 released by AnthropicMar 26OTIS Mock AIME 2024-2025 benchmark added
更多指标
Gecko Pulse
19+3
Pulse rose +3 this week · GPT-5.5 Pro drove benchmark velocity.
AI Bubble Index
278%↑
frothy · valuation premium leading the move.
Mindshare Pulse
78+12
Narrative energy spread across models · no single winner.
AICDI · Compute
71+2
Data-center intensity climbing · HBM bookings tight.
FCI · Foundries
62+1
TSMC N3 utilisation above 90% · queue deepening.
MDI · Memory
84+4
HBM3E scarcity driving GPU street prices up.
HBM Tension
68+3
Lead times stretching past 52 weeks · supply cliff watch.
Capex / Rev Gap
3.4×+0.2×
Hyperscaler capex outpacing AI revenue by 3.4× · widening.
矩阵 · 前 20 模型
| # | 基准测试 | ||||
|---|---|---|---|---|---|
| 1 | 99.9 | $30.00 | 400K | 3 | |
| 2 | 99.8 | — | 1000K | 14 | |
| 3 | 96.3 | $0.39 | 262K | 11 | |
| 4 | 95.2 | $0.40 | 164K | 9 | |
| 5 | 93.0 | $30.00 | 1050K | 8 | |
| 6 | 91.2 | $1.25 | 400K | 8 | |
| 7 | 90.0 | $2.00 | 1049K | 23 | |
| 8 | 89.5 | $0.10 | 262K | 10 | |
| 9 | 89.0 | $1.25 | 128K | 7 | |
| 10 | 88.7 | $0.33 | 1000K | 11 | |
| 11 | 88.3 | — | — | 11 | |
| 12 | HA Qwen2.5 72B Instruct Abliterated | 87.5 | — | — | 6 |
| 13 | 87.0 | $1.05 | 203K | 12 | |
| 14 | 85.4 | $1.75 | 400K | 9 | |
| 15 | 84.6 | — | — | 4 | |
| 16 | 84.4 | — | — | 7 | |
| 17 | 83.4 | $2.50 | 1050K | 16 | |
| 18 | 83.3 | $30.00 | 1000K | 12 | |
| 19 | 82.8 | $1.25 | 400K | 8 | |
| 20 | 81.7 | $0.09 | 262K | 11 |
技术栈 · 11 个数据层
模型
基准测试、价格、上下文 · 每个前沿模型
994 已追踪
供应商
跨所有 AI 推理供应商的可用性矩阵
268 已追踪
智能体
SWE-bench · MLE-bench · OSWorld 等
165 已追踪
基准测试
每项主要基准测试的分数演变
128 已追踪
研究
研究库 · 方法 · 笔记 · 来源地图
枢纽
经济
泡沫指数 · 估值 · 融资 · 市值
71 家公司
热度
Kaito 风格的全信号注意力追踪
93 个实体
基础设施
芯片 · 代工厂 · 内存 · 系统 · 能源
6 项指数
定价
套利发现 · 替代发现 · 免费层追踪
实时
学习
251 术语 · 词汇表 · 学习路径
251 术语
对比
通用对比 · 模型 · 智能体 · 供应商
15+ 路由
学习 AI
所有路径 →路径 · 7 术语
The AI Bubble Explained
Seven terms that decode whether AI is overpriced, fairly priced, or criminally underpriced. Read in order.
路径 · 6 术语
Pick an AI Model
Six terms to go from "I need an AI" to "here is the cheapest model that meets my spec."
路径 · 7 术语
From Sand to Model
The AI supply chain in 7 terms · foundry, memory, chip, system, datacenter, provider, API.
方法论 · FAQ · 新鲜度
完整方法论 →BenchGecko 数据多久更新一次？›
模型和基准测试数据每日从主要来源刷新。定价从每个供应商 API 滚动拉取。热度 信号每周汇总。脉搏 在 UTC 00:00 重新计算。
什么是 脉搏？›
一个 0-100 的 AI 经济健康综合分数。融合了反向 泡沫指数、基准测试速度、价格压缩、热度 多样性和供应链紧张度。数字越低越健康。
基准测试分数如何标准化？›
每项基准测试在所有已评分模型中进行最小最大归一化。排名取每个模型在 3 项以上基准测试中的归一化分数平均值，以避免过度加权单项测试。
定价数据来自哪里？›
直接来自供应商 API 响应 · OpenRouter、OpenAI、Anthropic、Google、xAI、DeepSeek、Mistral 等。每份快照都在模型详情页附有来源归属缓存。
可以引用 BenchGecko 的数据吗？›
可以。每个页面都提供分享与引用栏，含 APA、MLA、BibTeX、Chicago 和纯文本格式。免费 API 层要求归属，所有场景均建议归属。
数据来源 ·OpenRouterEpoch AISWE-benchMCP RegistryChatbot ArenaHuggingFaceLiveBenchArtificial AnalysisSEALAider