AI Arena

LLM Leaderboard

Live ranking of frontier language models by BenchLM composite score, with published API output pricing.

Source: BenchLM.ai · last verified November 19, 2026 · 214 ranked models tracked · top 30 shown here.
Data is CC BY-NC 4.0, © BenchLM.ai. This page is regenerated from their public leaderboard.

Score vs. cost

  • Proprietary (25)
  • Open Weight (4)
  • Pending (1)
  • Cost-efficient frontier
65 70 75 80 85 90 95 $1 $2 $5 $10 $20 $50 $100 $200 Output price per 1M tokens (USD, log scale) BenchLM score GPT-6.2 Astra91.1 · $40.00 Gemini 4 Neon86.3 · $15.00 GPT-6 Astra83.2 · $50.00 Claude Opus 5.582.7 · $20.00 GPT-6.2 Sol82.6 · $10.00 GPT-6.1 Sol80.7 · $10.00 Gemini 4 Argon79.1 · $10.00 Gemini 4 Hydrogen78.8 · $7.50 Claude Fable 5.178.7 · $50.00 Claude Sonnet 5.577.9 · $10.00 Claude Opus 574.9 · $25.00 Claude Fable 574.6 · $50.00 GPT-6 Sol73.3 · $10.00 GPT-5.6 Sol73.0 · $20.00 Muse Wonder71.2 · $20.00 Kimi K3.1 Pro70.6 · $10.00 GPT-5.5 Pro70.1 · $180.00 GPT-5.569.0 · $30.00 Hy468.5 · $2.501 GPT-5.4 Pro68.3 · $180.00 Gemini 3.8 Flash67.7 · $3.75 MiniMax M3.1 Flash68.1 · $0.60 DeepSeek V4.1 Flash 101769.9 · $1.20 Qwen4 Max74.5 · $4.50 GPT-6.1 Terra80.4 · $5.00 Gemini 4 Carbon88.1 · $10.00 Claude Fable 5.293.2 · $40.00 MiniMax M3.1 Flash DeepSeek V4.1 Flash 1017 Qwen4 Max GPT-6.1 Terra Gemini 4 Carbon Claude Fable 5.2
Higher and further left is better: the frontier joins models no other model beats on both score and price. 3 of the top 30 publish no list price and are omitted here; hover a point for its name and price, or read every value in the table below.

Top 30

#ModelProviderTypeContextScoreOutput $/1M
1Claude Fable 5.2AnthropicProprietary1M93.2$40.00
2GPT-6.2 AstraOpenAIProprietary1.05M91.1$40.00
3Gemini 4 CarbonGoogleProprietary1M88.1$10.00
4Gemini 4 NeonGoogleProprietary1M86.3$15.00
5GPT-6 AstraOpenAIProprietary1.05M83.2$50.00
6Claude Opus 5.5AnthropicProprietary1M82.7$20.00
7GPT-6.2 SolOpenAIProprietary1.05M82.6$10.00
8GPT-6.1 SolOpenAIProprietary1.05M80.7$10.00
9GPT-6.1 TerraOpenAIProprietary1.05M80.4$5.00
10Gemini 4 ArgonGoogleProprietary1M79.1$10.00
11Gemini 4 HydrogenGoogleProprietary1M78.8$7.50
12Claude Fable 5.1AnthropicProprietary1M78.7$50.00
13Claude Sonnet 5.5AnthropicProprietary1M77.9$10.00
14Muse Wonder FullMetaProprietary1M77.2—
15Claude Opus 5AnthropicProprietaryNot listed74.9$25.00
16Claude Fable 5AnthropicProprietary1M+74.6$50.00
17Qwen4 MaxAlibabaProprietary1M74.5$4.50
18GPT-6 SolOpenAIProprietary1.05M73.3$10.00
19GPT-5.6 SolOpenAIProprietary1.05M73.0$20.00
20Muse WonderMetaProprietary1M71.2$20.00
21Kimi K3.1 ProMoonshot AIOpen Weight1.05M70.6$10.00
22GPT-5.5 ProOpenAIProprietary1M70.1$180.00
23DeepSeek V4.1 Flash 1017DeepSeekOpen Weight1M69.9$1.20
24GPT-5.5OpenAIProprietary1M69.0$30.00
25Hy4TencentPending1M68.5$2.501
26GPT-5.4 ProOpenAIProprietary1.05M68.3$180.00
27Hy4 previewTencentOpen Weight1M68.2—
28MiniMax M3.1 FlashMiniMaxProprietary1M68.1$0.60
29Gemini 3.8 FlashGoogleProprietary1M67.7$3.75
30Ling 3.1 FlashInclusionAIOpen Weight262K66.8—

Reading the table

  • Score — BenchLM’s BenchAlign composite across their benchmark suite. BenchLM never interpolates missing scores, so a model’s score can be lower-confidence without being wrong.
  • Output $/1M — the provider’s list price per million output tokens. Input and cached prices are usually lower; check the model page before budgeting.
  • Context — maximum context window. M is a million tokens.
  • — — the provider does not publish a list price (typically self-hosted or open weights).

Notes

  • Ranked entries: 214. BenchLM separates Supported-ranked rows (exact source citations) from Estimated-ranked rows; lower ranks mix both.
  • Scores move between snapshots as new evaluations land, so treat close ranks as ties.
  • Regenerated automatically; the table above reflects the fetch on 2026-11-20.