Skip to content
e/acc model frontier · Consensus rankings from public benchmarks

Where is the frontier now?

Rankings draw on real public benchmarks across capabilities. Consensus indices are not accuracy, missing results are not zero, and prices do not affect ranks.
Time machineAcceleration

Financial analysis, legal advice and banking test whether professional tasks can be completed.

2 benchmarks·2 organizations·10/17 08:05New

Professional work boardTOP 30

Ranked by shared evidence across public benchmarks
Sortable top 30 models
Models
01Gemini 4 ArgonGoogleReleased2026-09-30 · Evidence-sensitiveCached input—Input—· Output—99.9
02Claude Sonnet 5.5AnthropicReleased— · Evidence-sensitiveCached input—Input—· Output—99.7
03Claude Opus 5.5AnthropicReleased2026-09-17 · Evidence-sensitiveCached input¥1.34Input¥26.74· Output¥133.7299.6
04Claude Fable 5.1AnthropicReleased2026-09-01 · Evidence-sensitiveCached input¥1.67Input¥66.86· Output¥334.3199.5
05Gemini 3.7 FlashGoogleReleased2026-08-13 · Evidence-sensitiveCached input¥0.5Input¥5.01· Output¥25.0799.2
06Claude Opus 5AnthropicReleased2026-07-24 · Evidence-sensitiveCached input¥3.34Input¥33.43· Output¥167.1698.4
07Grok 4.6xAIReleased2026-08-12 · Evidence-sensitiveCached input¥3.34Input¥13.37· Output¥40.1296.8
08GPT-6 AstraOpenAIReleased2026-09-03 · Evidence-sensitiveCached input¥6.69Input¥66.86· Output¥334.3195.4
09Gemini 3.8 FlashGoogleReleased2026-09-02 · Evidence-sensitiveCached input¥0.5Input¥5.01· Output¥25.0795.0
10Claude Fable 5AnthropicReleased2026-06-09 · Evidence-sensitiveCached input¥6.69Input¥66.86· Output¥334.3191.4
11Qwen3.8 MaxAlibabaReleased2026-07-19 · Evidence-sensitiveCached input—Input¥12· Output¥3686.8
12GPT-6.1 SolOpenAIReleased— · Evidence-sensitiveCached input—Input—· Output—86.0
13GPT-5.6 TerraOpenAIReleased2026-07-09 · Evidence-sensitiveCached input¥1.34Input¥13.37· Output¥80.2384.7
14GLM-5.3Z.aiReleased2026-08-18 · Evidence-sensitiveCached input¥1.74Input¥9.36· Output¥29.4282.6
15Grok 4.5xAIReleased2026-07-08 · Evidence-sensitiveCached input¥2.01Input¥13.37· Output¥40.1277.5
16Claude Haiku 5.5AnthropicReleased2026-10-07 · Evidence-sensitiveCached input—Input—· Output—76.7
17DeepSeek V4 Flash 0731DeepSeekReleased2026-07-31 · Evidence-sensitiveCached input¥0.1Input¥3· Output¥971.7
18GPT-5.5OpenAIReleased2026-04-23 · Evidence-sensitiveCached input¥3.34Input¥33.43· Output¥200.5971.0
19Grok 4.7xAIReleased2026-09-21 · Evidence-sensitiveCached input¥3.34Input¥13.37· Output¥40.1269.3
20Claude Sonnet 5AnthropicReleased2026-06-30 · Evidence-sensitiveCached input¥1.34Input¥13.37· Output¥66.8668.5
21MiMo V 2.6 ProXiaomiReleased2026-09-21 · Evidence-sensitiveCached input—Input—· Output—67.8
22GPT-6 SolOpenAIReleased2026-09-22 · Evidence-sensitiveCached input¥1.34Input¥13.37· Output¥66.8662.6
23GLM-5.3 FlashZ.aiReleased2026-08-26 · Evidence-sensitiveCached input¥0.23Input¥0.8· Output¥2.861.8
24GPT-5.6 SolOpenAIReleased2026-07-09 · Evidence-sensitiveCached input¥2.67Input¥26.74· Output¥133.7256.5
25Claude Opus 4.7AnthropicReleased2026-04-16 · Evidence-sensitiveCached input¥3.34Input¥33.43· Output¥167.1650.6
26Claude Opus 4.8AnthropicReleased2026-05-28 · Evidence-sensitiveCached input¥3.34Input¥33.43· Output¥167.1649.8
27Kimi K3Moonshot AIReleased2026-07-16 · Evidence-sensitiveCached input¥2Input¥20· Output¥10044.0
28Muse Spark 1.3MetaReleased2026-09-02 · Evidence-sensitiveCached input—Input¥8.36· Output¥28.4243.2
29Qwen3.8 27 BAlibabaReleased2026-08-14 · Evidence-sensitiveCached input¥0.6Input¥3· Output¥1237.7
30DeepSeek V4 Pro 0813DeepSeekReleased2026-08-13 · Evidence-sensitiveCached input¥0.3Input¥9· Output¥2736.1

Up to 30 models per board

The consensus index is not accuracy. Tied scores retain ranks determined by the full evidence.

How to read this board

Currently covers financial analysis, professional consulting and banking. It does not yet represent all document, spreadsheet and presentation tasks.Prices do not affect rank, missing results are not zero, and the index is not accuracy.

Read the methodology →

About prices

API prices come from vendor websites and are shown per million tokens.USD prices are converted to CNY at 2026-10-16.Cached prices cover cache-hit input. Cache writes, storage and subscriptions are separate.

How fast the same capability gets cheaper: the cost of intelligence →