BoilingFrogs logo
BoilingFrogs logo AI literacy Latest briefings About the mission

Current Leading Models by Task

Last Refresh 6 Jul 2026, 06:02 UTC

General text-to-text inference

Efficacy score blends key benchmark rankings for this task type.

# Move Model Provider Open? 🌐 Cost Efficacy
#1 → 0 Claude Fable 5 (with fallback)AA Intelligence Index: 59.86 from artificial analysis llm. Anthropic Closed 🇺🇸 Premium $20.00 blended / 1M tokens 98
#2 → 0 Claude Opus 4.8 (max)AA Intelligence Index: 55.69 from artificial analysis llm. Anthropic Closed 🇺🇸 Premium $10.00 blended / 1M tokens 92
#3 → 0 GPT-5.5 (xhigh)AA Intelligence Index: 54.84 from artificial analysis llm. OpenAI Closed 🇺🇸 Premium $11.25 blended / 1M tokens 91
#4 → 0 Claude Opus 4.7 (max)AA Intelligence Index: 53.53 from artificial analysis llm. Anthropic Closed 🇺🇸 Premium $10.00 blended / 1M tokens 89
#5 → 0 Claude Sonnet 5 (max)AA Intelligence Index: 53.35 from artificial analysis llm. Anthropic Closed 🇺🇸 Premium $6.00 blended / 1M tokens 89
#6 → 0 GPT-5.5 (high)AA Intelligence Index: 53.13 from artificial analysis llm. OpenAI Closed 🇺🇸 Premium $11.25 blended / 1M tokens 88
#7 → 0 GPT-5.4 (xhigh)AA Intelligence Index: 51.40 from artificial analysis llm. OpenAI Closed 🇺🇸 Premium $5.63 blended / 1M tokens 86
#8 → 0 GLM-5.2 (max)AA Intelligence Index: 51.09 from artificial analysis llm. Z AI Open 🇨🇳 High $2.15 blended / 1M tokens 85
#9 → 0 GPT-5.5 (medium)AA Intelligence Index: 50.41 from artificial analysis llm. OpenAI Closed 🇺🇸 Premium $11.25 blended / 1M tokens 85
#10 → 0 Gemini 3.5 FlashAA Intelligence Index: 50.20 from artificial analysis llm. Google Closed 🇺🇸 Premium $3.38 blended / 1M tokens 84
#11 → 0 Claude Sonnet 4.6 (max)AA Intelligence Index: 47.21 from artificial analysis llm. Anthropic Closed 🇺🇸 Premium $6.00 blended / 1M tokens 80
#12 → 0 Gemini 3.1 Pro PreviewAA Intelligence Index: 46.46 from artificial analysis llm. Google Closed 🇺🇸 Premium $4.50 blended / 1M tokens 79
#13 → 0 Qwen3.7 MaxAA Intelligence Index: 45.99 from artificial analysis llm. Alibaba Closed 🇨🇳 Premium $3.75 blended / 1M tokens 78
#14 → 0 Gemini 3.5 Flash (medium)AA Intelligence Index: 45.38 from artificial analysis llm. Google Closed 🇺🇸 Premium $3.38 blended / 1M tokens 77
#15 → 0 MiniMax-M3AA Intelligence Index: 44.44 from artificial analysis llm. MiniMax Open 🇨🇳 Low $0.52 blended / 1M tokens 76
#16 → 0 DeepSeek V4 Pro (max)AA Intelligence Index: 44.27 from artificial analysis llm. DeepSeek Open 🇨🇳 Low $0.54 blended / 1M tokens 76
#17 → 0 GPT-5.3 Codex (xhigh)AA Intelligence Index: 44.27 from artificial analysis llm. OpenAI Closed 🇺🇸 Premium $4.81 blended / 1M tokens 76
#18 → 0 Claude Opus 4.6 (max)AA Intelligence Index: 43.71 from artificial analysis llm. Anthropic Closed 🇺🇸 Premium $10.00 blended / 1M tokens 75
#19 → 0 GPT-5.5 (low)AA Intelligence Index: 43.46 from artificial analysis llm. OpenAI Closed 🇺🇸 Premium $11.25 blended / 1M tokens 75
#20 → 0 Muse SparkAA Intelligence Index: 43.06 from artificial analysis llm. Meta Closed 🇺🇸 High not consistently published 74

Method

How the efficacy rating is built

  • Each refresh scrapes public benchmark pages/data sources, stores the raw/parsed scores in PostgreSQL, then recomputes each task leaderboard.
  • Efficacy is a 0–100 normalisation within each task category: the top parsed benchmark score is near the high 90s and followers are scaled beneath it.
  • Movement compares today’s PostgreSQL rank with the previous persisted run for the same task and model.
  • Cost, open/closed status and provider country are context columns only; they do not raise or lower efficacy.
  • The benchmark catalogue below lists the wider credible source pool, including boards that are relevant but do not publish easy structured per-model data.

General text-to-text inference

11 public task-relevant leaderboards / benchmark boards

  1. Artificial Analysis LLM Leaderboard Commercial/open LLM intelligence, coding, agentic, speed and price indices.
  2. LMArena Chatbot Arena Human preference comparisons for chat, style, vision and image arenas.
  3. LiveBench Contamination-resistant reasoning, math, coding, language and data-analysis tasks.
  4. OpenCompass LLM Leaderboard Broad multi-benchmark LLM evaluation across open and closed models.
  5. Stanford HELM Lite Standardised model accuracy, calibration, robustness and efficiency.
  6. Hugging Face Open LLM Leaderboard Open model benchmarks and reproducible community evaluation.
  7. SEAL Leaderboards Private expert exams for frontier model intelligence and domain ability.
  8. SimpleBench Short hard reasoning questions designed to expose brittle reasoning.
  9. GPQA Leaderboard / benchmark Graduate-level science question answering.
  10. WildBench Open-ended real-user prompts including writing, style and instruction following.
  11. Arena-Hard-Auto Hard open-ended assistant tasks; proxy for polished long-form responses.

Coding

11 public task-relevant leaderboards / benchmark boards

  1. Artificial Analysis LLM Leaderboard Commercial/open LLM intelligence, coding, agentic, speed and price indices.
  2. LiveBench Contamination-resistant reasoning, math, coding, language and data-analysis tasks.
  3. OpenCompass LLM Leaderboard Broad multi-benchmark LLM evaluation across open and closed models.
  4. Hugging Face Open LLM Leaderboard Open model benchmarks and reproducible community evaluation.
  5. SEAL Leaderboards Private expert exams for frontier model intelligence and domain ability.
  6. SWE-bench Verified Real GitHub issue resolution with verified tests.
  7. LiveCodeBench Recent programming problems and code-generation pass rates.
  8. Aider Polyglot Leaderboard Practical repository editing across multiple languages.
  9. BigCodeBench Practical function-level software engineering tasks.
  10. EvalPlus HumanEval+/MBPP+ code correctness under stronger tests.
  11. Terminal-Bench Terminal-based agent tasks and command-line problem solving.

Creative writing

6 public task-relevant leaderboards / benchmark boards

  1. LMArena Chatbot Arena Human preference comparisons for chat, style, vision and image arenas.
  2. EQ-Bench Creative Writing v3 LLM prose quality, voice, style and creative-writing judgement.
  3. EQ-Bench Longform Writing Long-form writing and coherence assessment.
  4. WritingBench Professional and creative writing instruction-following.
  5. WildBench Open-ended real-user prompts including writing, style and instruction following.
  6. Arena-Hard-Auto Hard open-ended assistant tasks; proxy for polished long-form responses.

Computer use

8 public task-relevant leaderboards / benchmark boards

  1. Artificial Analysis LLM Leaderboard Commercial/open LLM intelligence, coding, agentic, speed and price indices.
  2. Terminal-Bench Terminal-based agent tasks and command-line problem solving.
  3. OSWorld Desktop computer-control tasks across real operating-system apps.
  4. WebArena Realistic browser tasks over websites and web apps.
  5. VisualWebArena Visual browser-use tasks requiring perception plus action.
  6. WorkArena Enterprise workflow automation in ServiceNow-style environments.
  7. Online-Mind2Web Live web navigation and browser-agent generalisation.
  8. BrowserGym Unified benchmark harness for web agents.

Image generation

7 public task-relevant leaderboards / benchmark boards

  1. Artificial Analysis Text-to-Image Leaderboard Text-to-image quality Elo, speed and price.
  2. LMArena Chatbot Arena Human preference comparisons for chat, style, vision and image arenas.
  3. GenEval Compositional text-to-image object, count, colour and relation following.
  4. T2I-CompBench Fine-grained compositional prompt-following for text-to-image systems.
  5. DPG-Bench Dense-prompt adherence for generated images.
  6. HPD v2 Human preference scoring for image generation.
  7. PartiPrompts Prompt suite for text-to-image compositional and stylistic coverage.

Source pool

Credible benchmark boards tracked

32 public leaderboards and benchmark boards checked or catalogued for the daily refresh.

Benchmark Owner / site Specialises in Areas
Artificial Analysis LLM Leaderboard Artificial Analysis Commercial/open LLM intelligence, coding, agentic, speed and price indices. text, coding, computer
Artificial Analysis Text-to-Image Leaderboard Artificial Analysis Text-to-image quality Elo, speed and price. image-generation
LMArena Chatbot Arena LMArena / LMSYS Human preference comparisons for chat, style, vision and image arenas. text, writing, image-generation
LiveBench LiveBench Contamination-resistant reasoning, math, coding, language and data-analysis tasks. text, coding
OpenCompass LLM Leaderboard OpenCompass / Shanghai AI Laboratory Broad multi-benchmark LLM evaluation across open and closed models. text, coding
Stanford HELM Lite Stanford CRFM Standardised model accuracy, calibration, robustness and efficiency. text
Hugging Face Open LLM Leaderboard Hugging Face Open model benchmarks and reproducible community evaluation. text, coding
SEAL Leaderboards Scale AI Private expert exams for frontier model intelligence and domain ability. text, coding
SimpleBench SimpleBench Short hard reasoning questions designed to expose brittle reasoning. text
GPQA Leaderboard / benchmark NYU / Cohere / Anthropic authors Graduate-level science question answering. text
SWE-bench Verified SWE-bench Real GitHub issue resolution with verified tests. coding
LiveCodeBench LiveCodeBench Recent programming problems and code-generation pass rates. coding
Aider Polyglot Leaderboard Aider Practical repository editing across multiple languages. coding
BigCodeBench BigCode / Salesforce Research Practical function-level software engineering tasks. coding
EvalPlus EvalPlus HumanEval+/MBPP+ code correctness under stronger tests. coding
Terminal-Bench Terminal-Bench Terminal-based agent tasks and command-line problem solving. coding, computer
EQ-Bench Creative Writing v3 EQ-Bench LLM prose quality, voice, style and creative-writing judgement. writing
EQ-Bench Longform Writing EQ-Bench Long-form writing and coherence assessment. writing
WritingBench X-PLUG / academic authors Professional and creative writing instruction-following. writing
WildBench AllenAI Open-ended real-user prompts including writing, style and instruction following. text, writing
Arena-Hard-Auto LMSYS Hard open-ended assistant tasks; proxy for polished long-form responses. text, writing
OSWorld OSWorld academic authors Desktop computer-control tasks across real operating-system apps. computer
WebArena CMU / WebArena authors Realistic browser tasks over websites and web apps. computer
VisualWebArena VisualWebArena authors Visual browser-use tasks requiring perception plus action. computer
WorkArena ServiceNow Research Enterprise workflow automation in ServiceNow-style environments. computer
Online-Mind2Web OSU NLP Group Live web navigation and browser-agent generalisation. computer
BrowserGym ServiceNow / browser-agent community Unified benchmark harness for web agents. computer
GenEval GenEval authors Compositional text-to-image object, count, colour and relation following. image-generation
T2I-CompBench T2I-CompBench authors Fine-grained compositional prompt-following for text-to-image systems. image-generation
DPG-Bench Tencent / ELLA project Dense-prompt adherence for generated images. image-generation
HPD v2 Human Preference Dataset authors Human preference scoring for image generation. image-generation
PartiPrompts Google Research Prompt suite for text-to-image compositional and stylistic coverage. image-generation