OverviewAll LeaderboardsMethodologyComing soon
New Chat

Chat Index

What is the best agentic chat model?

Nov 13, 2026
962,655 sessions
65 models

Chat Index

Frontier
AnthropicClaude Fable 5.2 (Max)1,386 on the index. 65 points above rank 2.
Daily Driver
Gemini 4 Argon (High)Near-frontier performance at 1/10th the price.
Efficient
MetaMuse Ember (Max)Solid performance at 1/100th the price.
1,0001,1001,2001,3001,400
  1. 1,386
    Anthropic
    Claude Fable 5.2Max
  2. 1,321
    Anthropic
    Claude Opus 5.5Max
  3. 1,315
    Gemini 4 HydrogenHighPreliminary
  4. 1,304
    Gemini 4 CarbonHighPreliminary
  5. 1,294
    Anthropic
    Claude Sonnet 5.5Max
  6. 1,263
    Anthropic
    Claude Fable 5.1Max
  7. 1,261
    GPT-6 AstraMax
  8. 1,238
    Anthropic
    Claude Opus 5.5High
  9. 1,235
    Gemini 4 ArgonHigh
  10. 1,202
    GPT-6.1 SolMax
  11. 1,202
    GPT-6.1 TerraMax
  12. 1,179
    Meta
    Muse Wonder FullMax
  13. 1,175
    GPT-6.1 SolMedium
  14. 1,171
    Qwen4-MaxxHighPreliminary
  15. 1,164
    Anthropic
    Claude Opus 5Max
Model RankingIndexCost per task (P50)Time per task (P50)
  1. 1AnthropicClaude Fable 5.2 (Max)1,386$5.5510.8 min
  2. 2AnthropicClaude Opus 5.5 (Max)1,321$4.1311.6 min
  3. 3Gemini 4 Hydrogen (High)Preliminary1,315$1.033.4 min
  4. 4Gemini 4 Carbon (High)Preliminary1,304$0.845.7 min
  5. 5AnthropicClaude Sonnet 5.5 (Max)1,294$3.0515.2 min
  6. 6AnthropicClaude Fable 5.1 (Max)1,263$2.724 min
  7. 7GPT-6 Astra (Max)1,261$1.904.5 min
  8. 8AnthropicClaude Opus 5.5 (High)1,238$1.152.6 min
  9. 9Gemini 4 Argon (High)1,235$0.474.7 min
  10. 10GPT-6.1 Sol (Max)1,202$0.504.2 min
  11. 11GPT-6.1 Terra (Max)1,202$0.636.1 min
  12. 12MetaMuse Wonder Full (Max)1,179N/AN/A
  13. 13GPT-6.1 Sol (Medium)1,175$0.281.4 min
  14. 14Qwen4-Max (xHigh)Preliminary1,171$0.839.3 min
  15. 15AnthropicClaude Opus 5 (Max)1,164$2.445.2 min

Pareto frontier

Best index for the cost
9001,0001,1001,2001,3001,400
MetaAnthropicAnthropic
$0.01$0.10$1$10

Median cost per task (last 14 days)

Frontier
AnthropicClaude Fable 5.2 (Max)1,386 on the index. 65 points above rank 2.
Daily Driver
Gemini 4 Argon (High)Near-frontier performance at 1/10th the price.
Efficient
MetaMuse Ember (Max)Solid performance at 1/100th the price.
On the frontierIndex$/task
  1. AnthropicClaude Fable 5.2 (Max)1,386$5.55
  2. AnthropicClaude Opus 5.5 (Max)1,321$4.13
  3. Gemini 4 Hydrogen (High)1,315$1.03
  4. Gemini 4 Carbon (High)1,304$0.84
  5. Gemini 4 Argon (High)1,235$0.47
  6. GPT-6.1 Sol (Medium)1,175$0.28
  7. GPT-6.1 Terra (Medium)1,117$0.14
  8. MetaMuse Ember (Max)1,105$0.03
  9. GPT-6 Luna (Medium)1,023$0.02
  10. GPT-6 Luna (Low)961$0.01
7 models not plotted

These models are on the leaderboard, but they have too few sessions with a measured cost for a point on this chart. The numbers beside their names show their ranks.

  • 12Muse Wonder Full (Max)
  • 45Beam 1.0 501B A23B
  • 47Muse Spark 1.1
  • 56Mistral Medium 3.5
  • 61Inkling
  • 62NVIDIA-Nemotron-3.5-Super-VL-120B-A12B (NVFP4)
  • 64Inkling-Small

Signal Performance

Ranking by Task CompletionScore
  1. 1AnthropicClaude Opus 5.5 (Max)1,370
  2. 2AnthropicClaude Fable 5.2 (Max)1,365
  3. 3Gemini 4 Hydrogen (High)1,320
  4. 4Gemini 4 Carbon (High)1,311
  5. 5AnthropicClaude Sonnet 5.5 (Max)1,296
  6. 6AnthropicClaude Fable 5.1 (Max)1,286
  7. 7GPT-6 Astra (Max)1,273
  8. 8Gemini 4 Argon (High)1,270
  9. 9AnthropicClaude Opus 5.5 (High)1,257
  10. 10MetaMuse Wonder Full (Max)1,227
  11. 11GPT-6.1 Sol (Max)1,222
  12. 12Qwen4-Max (xHigh)1,211
  13. 13GPT-6.1 Terra (Max)1,197
  14. 14AnthropicClaude Sonnet 5.5 (High)1,196
  15. 15Gemini 3.8 Flash (High)1,182

Each rank counts all 65 models · Each bar starts at 1,100 and ends at 1,400 · Each signal is on the scale of the index

Overall Ranking by Index
Index
Task Completion
Praise vs Complaint
Steerability
Bash Reliability

Each signal rank counts every model of the ranking, not only the models in this card

Previous
Coding Index

Which agents are best at real coding work?

Next
Task Outcome Index

Which model is the best at completing user tasks?

Frequently asked questions

Agent Mode

Try Agent Mode

Put these models to work on your own real tasks in Agent Mode.

Get started
How the Agent Leaderboard works

How the Agent Leaderboard works

See how we turn millions of real Agent Mode sessions into causal, per-signal scores.

Read the methodology

USE CASES

  • Chat with AI
  • Build Apps & Websites
  • Write & Edit Text
  • Search the Web
  • Generate Images
  • Generate Videos
  • Chose any model
  • Compare Models Side by Side

LEADERBOARD RANKINGS

  • Overall
  • Agent
  • Alignment Index
  • Text
  • WebDev
  • Image-to-WebDev
  • Text to Image
  • Image Edit
  • Text to Video
  • Image to Video
  • Video Edit
  • Vision
  • Document
  • Search

COMPANY

  • About Us
  • How It Works
  • Blog
  • Careers
  • Leaderboard Changelog
  • Product Changelog
  • Help Center
  • FAQ

LEGAL

  • Terms
  • Privacy
  • Cookies

FOLLOW

  • X
  • LinkedIn
  • YouTube
  • Discord

© Arena Intelligence