XBench

Home = our average · lab pages = their ranks

As of 2026-10-08

BenchLM.

Their board. Their ranks.

BenchLM hosts AA Agentic and Coding mirrors plus SWE-bench Verified and SWE-bench Pro. Rank follows SWE-Pro when present, else AA Coding.

BenchLM #1 is Opus 5.5

Open the lab’s original

BenchLM

Aggregator snapshot, 23 Sep 2026. AA mirrors are the same numbers as the AA page — SWE boards are not.

1
Opus 5.5
Anthropic · closed
Anthropic’s 5.5 frontier flagship
89.9%Roundup———
2
Fable 5.1
Anthropic · closed
New Claude 5 frontier tier
81.2%Roundup—58AA Agentic81.6AA Coding
3
Mythos 5
Anthropic · closed
New Claude 5 frontier tier
80.3%Roundup95.5%Roundup88%TB 2.080.3%SWE-Pro
4
Fable 5
Anthropic · closed
New Claude 5 frontier tier
80%Roundup95%Roundup51AA Agentic76.5AA Coding
5
Opus 5
Anthropic · closed
Anthropic’s flagship frontier model
79.2%Roundup96%Roundup56.2AA Agentic78AA Coding
19
Opus 4.8
Anthropic · closed
Frontier reasoning/coding
69.2%Roundup88.6%Roundup42.6AA Agentic74.3AA Coding
23
Hy4
Tencent · open
Open Apache-2.0 coding/reasoning
65.7%Roundup———
24
GPT-5.6 Sol
OpenAI · closed
Top-tier flagship reasoning
64.6%Roundup—50.5AA Agentic77.4AA Coding
25
Opus 4.7
Anthropic · closed
Frontier coding/reasoning
64.3%Roundup87.6%Roundup39.5AA Agentic73.6AA Coding
26
Sonnet 5
Anthropic · closed
Frontier general-purpose/agentic
63.2%Roundup85.2%Roundup44.3AA Agentic71.5AA Coding
27
Qwen3.7 Max
Alibaba · closed
Frontier Chinese model
60.6%Roundup80.4%Roundup23.9AA Agentic66AA Coding
29
GPT-5.5
OpenAI · closed
Frontier reasoning + agents
58.6%Roundup—37.3AA Agentic74.9AA Coding
31
MiMo V2.5 Pro
Xiaomi · open
Open MIT, cheap per-task
57.2%Roundup—22.7AA Agentic60.2AA Coding
32
V4 Pro 0813
DeepSeek · open
Current DeepSeek-V4 open challenger
55.4%Roundup80.6%Roundup49.6AA Agentic68.8AA Coding
33
Inkling
Thinking Machines · open
New open lab, Apache-2.0
54.3%Roundup77.6%Roundup24.3AA Agentic52.1AA Coding
34
V4 Flash
DeepSeek · open
Open price-performance pick
52.6%Roundup79%Roundup41.7AA Agentic69.1AA Coding
35
Grok 4.20
SpaceXAI · closed
Frontier reasoning/agents
51.8%Roundup76.7%Roundup47.1%TB 2.076.7%SWE Verified
6
GPT-6 Astra
OpenAI · closed
OpenAI GPT-6 flagship
——51.5AA Agentic76.9AA Coding
7
Grok 4.6
SpaceXAI · closed
Latest Grok frontier model
——53.4AA Agentic76.8AA Coding
10
Kimi K3
Moonshot AI · open
Frontier open-weight
——50.6AA Agentic76.2AA Coding
17
Qwen3.8 Max
Alibaba · closed
Hosted 2.4T flagship; weights promised
——49.6AA Agentic71.8AA Coding
14
Flash Next
Alibaba · open
Qwen4-architecture open preview
———73AA Coding
16
Spark 1.2
Meta · closed
Agent-native Meta model
——44AA Agentic72.2AA Coding
12
Spark 1.3
Meta · closed
Agent-native Meta model; reaches the frontier
——55.7AA Agentic75.8AA Coding
8
GPT-5.6 Terra
OpenAI · closed
Balanced GPT-5.6
——43.7AA Agentic76.7AA Coding
9
3.8 Flash
Google DeepMind · closed
Latest Flash workhorse
——41.1AA Agentic76.3AA Coding
11
3.7 Flash
Google DeepMind · closed
Latest Flash workhorse
——36.4AA Agentic76.1AA Coding
15
Grok 4.5
SpaceXAI · closed
Frontier coding/agents
——42.1AA Agentic72.5AA Coding
13
GLM-5.3
Z.ai · staged
Latest GLM coding / cyber-defense
——53.4AA Agentic74.8AA Coding
21
GLM-5.2
Zhipu AI · open
Frontier/open-weight coding
——39.4AA Agentic68.8AA Coding
18
GPT-5.6 Luna
OpenAI · closed
Cost-efficient GPT-5.6
——42.7AA Agentic71.5AA Coding
19
3.6 Flash
Google DeepMind · closed
Frontier-class fast reasoning
——30.1AA Agentic69.2AA Coding
21
3.1 Pro
Google DeepMind · closed
Frontier multimodal/reasoning
——10.3AA Agentic68.8AA Coding
36
Grok 4.3
SpaceXAI · closed
Frontier reasoning/agents
——17.2AA Agentic42.3AA Coding
28
Hy3
Tencent · open
Open Apache-2.0 coding/reasoning
——25.6AA Agentic58.8AA Coding
29
MiniMax M3
MiniMax · open
Open-weight frontier-adjacent
—80.5%Roundup30.8AA Agentic58.6AA Coding

36 models on this board · # is this page’s rank