Part of Best LLM for AI Agents

Best LLM for Tool & Function Calling

Claude Fable 5.2 is the provisional first pick for Tool & Function Calling as of November 15, 2026, while independent task measurements are still limited.[1][2][3][4] Ranks #1 on LMArena's agentic arena (score 30.4) and achieves perfect tool-hallucination and steerability scores, indicating strong multi-step tool-use performance with minimal fabricated calls. GPT-6 Astra is the next-ranked alternative. It ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.

About this recommendation

Updated
Evidence through
Sources
13
Revision
v137

Decision audit

Why this result

Inspect the inputs and the computed order behind the recommendation.

Models screened

24

live candidates

Evaluation feeds

7

task-weighted

Winner coverage

38%

intended feed weight

Largest provider share

3 of 8

Anthropic

Provisional source breadth. 3 citation families and 0 practitioner families provide attributed capability reports; 0 cautionary threads retained. The largest citation family contributes 89%.

Anthropic: Claude Fable 5.2 leads the measured task comparison with limited independent task measurements. Community volume, release age and usage popularity add no ranking points.

Real-world track record is still developing. The recommendation can lead on independent measurements while practitioner reports accumulate.

Sources evaluated

These are the weights used in scoring. Missing measurements are omitted from each model’s average. A feed with no result for any model released in the last 60 days is stale and does not score until it catches up.

winner: Claude Fable 5.2
Evaluation feedWeightWinner resultField measured
Berkeley Function Callingstale
0%
no current result0/24
tau2-benchstale
0%
no current result0/24
LMArena Agent Tool HallucinationObserved 2026-11-11 · evaluation date not supplied · max effort
36%
#123/24
LMArena AgentObserved 2026-11-11 · evaluation date not supplied · max effort
27%
#123/24
LMArena Agent SteerabilityObserved 2026-11-11 · evaluation date not supplied · max effort
18%
#123/24
LMArena Search
18%
not measured5/24
Structured-output evaldisabled; last successful check unknownunavailable
0%
feed unavailable0/24

Provider concentration

Each exact model is scored separately; provider identity is not a ranking input.

Anthropic38%
  • Anthropic3 models
  • OpenAI3 models
  • Google1 model
  • xAI1 model

Decision table

Every published model is shown in computed order. Practitioner sources are distinct community threads, not the citations repeated in the prose below.

  1. Anthropic · coverage 38% · provisional evidence

    #1 LMArena Agent · #1 LMArena Agent Steerability

  2. OpenAI · coverage 46% · provisional evidence

    #1 LMArena Agent Tool Hallucination · #4 LMArena Search

  3. Anthropic · coverage 46% · provisional evidence

    #2 LMArena Agent · #3 LMArena Agent Steerability

  4. xAI · coverage 8% · provisional evidence

    #8 LMArena Search

  5. OpenAI · coverage 46% · provisional evidence

    #1 LMArena Agent Tool Hallucination · #5 LMArena Search

  6. Anthropic · coverage 38% · provisional evidence

    #4 LMArena Agent · #7 LMArena Agent Steerability

  7. Google · coverage 38% · provisional evidence

    #2 LMArena Agent Steerability · #3 LMArena Agent

  8. OpenAI · coverage 38% · provisional evidence

    #1 LMArena Agent Tool Hallucination · #4 LMArena Agent Steerability

Task score averages the available, task-weighted measurements. Benchmark positions compare models matched to our catalog. Models measured by at least two independent evaluator families are considered first. Missing tests, practitioner report counts, popularity and release age add no points. Coverage describes evidence breadth, not a probability of correctness. Scores within two points are shown as close under this policy; this is not a statistical confidence interval. When the leading scores are that close, the model measured by more evaluator families ranks first.

  1. 01

    Claude Fable 5.2Anthropic · $8 in / $40 out per M tokens · 1M context

    Ranks #1 on LMArena's agentic arena (score 30.4) and achieves perfect tool-hallucination and steerability scores, indicating strong multi-step tool-use performance with minimal fabricated calls.

    Best when: Use for multi-step agentic workflows where following mid-task direction matters, as it ranks #1 on LMArena Agent's steerability signal (score 33).
    Source 1
    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's steerability signal (score 33), measuring how well the agent follows direction given mid-task.”
    LMArena Agent steerabilityOpen original ↗

    More tips

    • Deploy when minimizing fabricated tool calls is critical, as it ranks #1 on LMArena Agent's tool-hallucination signal (score 1).
      Source 2
      “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.”
      LMArena Agent tool hallucinationOpen original ↗

    No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.

  2. 02

    GPT-6 AstraOpenAI · $10 in / $50 out per M tokens · 1.1M context

    GPT-6 Astra ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.

    Best when: Its strongest result for this task is on LMArena Agent tool hallucination, which measures calling only tools that exist, with real arguments.
    Source 4
    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.”
    LMArena Agent tool hallucinationOpen original ↗
  3. 03

    Claude Opus 5.5Anthropic · $4 in / $20 out per M tokens · 1M context

    Ranks #2 on LMArena's agentic arena (score 19.5) with strong steerability (#3, score 18.6).

    Best when: Use for agentic tasks requiring strong mid-task direction following, as it ranks #3 on LMArena Agent's steerability signal (score 18.6).
    Source 5
    “Within the 52 catalog models matched to this evaluation: Ranks #3 of 52 on LMArena Agent's steerability signal (score 18.6), measuring how well the agent follows direction given mid-task.”
    LMArena Agent steerabilityOpen original ↗

    Watch out for

    • Avoid for kernel-verified proof construction, as it failed to construct proofs in 12 baseline runs on 'impossible to pass' levels.
      Source 6
      “2026-10-29 清华姚期智团队提出 AWM:给陌生世界写定律并用 Lean 4 证明 - 清华交叉信息研究院姚期智、袁洋团队 10 月 29 日发布《Algebraic World Model》:把未知环境表示为「类型化对象 + 共享状态转换规则 + 目标条件」的显式数学模型,而非预测下一帧的神经网络。 - 流程三步:在环境中写可执行 Python 假设并用观测持续修正;用范畴论投射/归纳极限做结构化简;最后形式化为 Lean 4,证明目标可达或不可达。 - 两个「不可能通关」关卡:AWM 六次运行全部构造出内核检验通过的证明;配代码工具的 Claude Opus 5.5 与 GPT-6 Astra 在 12 次基线运行中一次都没做到。 - ARC-AGI-3 全部 25 个游戏、183 个关卡可让 Lean 复放通关轨迹;搭配 Opus 5 得分 99.9,获胜轨迹只用人类基线 43% 的动作。 - 相关链接:arXiv: [Algebraic World Model: From Exploration to P…”
  4. 04

    Grok 4.5xAI · $2 in / $6 out per M tokens · 500K context

    Grok 4.5 ranks #8 of 15 on LMArena's search arena (score 1216), measuring answers produced with search.

    Best when: Its strongest result for this task is on LMArena Search, which measures answers grounded in live search.
    Source 7
    “Within the 15 catalog models matched to this evaluation: Ranks #8 of 15 on LMArena's search arena (score 1216), measuring answers produced with search.”
    LMArena search arenaOpen original ↗
  5. 05

    GPT-6.1 SolOpenAI · $2 in / $10 out per M tokens · 1.1M context

    Ties for #1 on tool-hallucination signal with moderate agentic arena ranking (#7, score 12.5).

    Best when: Use when minimal tool hallucination is critical, as it ranks #1 on LMArena Agent's tool-hallucination signal (score 1).
    Source 8
    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.”
    LMArena Agent tool hallucinationOpen original ↗

    More tips

    • Deploy for search-augmented answers where it ranks #5 on LMArena's search arena (score 1230).
      Source 9
      “Within the 15 catalog models matched to this evaluation: Ranks #5 of 15 on LMArena's search arena (score 1230), measuring answers produced with search.”
      LMArena search arenaOpen original ↗
  6. 06

    Claude Sonnet 5.5Anthropic · $2 in / $10 out per M tokens · 1M context

    Solved 24 of 24 agent task benchmark runs with moderate LMArena rankings (#4 agentic arena, score 16; #7 steerability, score 14.5).

    Best when: Use for agent task benchmarks where it achieved perfect 24/24 success rate in controlled testing.
    Source 10
    “The agent task benchmark (docs/PERFORMANCE.md → "Agent task benchmark") solved 24 of 24 runs on Claude Sonnet 5.5, so it can no longer show where the agent struggles or which change made it better. Record more per run (in tests/bench/agent_tasks.bench.ts): - model requests (#42); - tool calls by name (read_file, grep, edit_file, run_command, …), to see how a task was approached; - the reason for each failed call (see the issue on failed tool cards); - whether files changed without…”

    More tips

    • Deploy for multi-step tool-use workflows, ranking #4 on LMArena's agentic arena (score 16).
      Source 11
      “Within the 52 catalog models matched to this evaluation: Ranks #4 of 52 on LMArena's agentic arena (score 16), measuring tool-use and multi-step task performance from human preference.”
      LMArena agentic arenaOpen original ↗
  7. 07

    Gemini 4 CarbonGoogle · $2 in / $10 out per M tokens · 1.0M context

    Gemini 4 Carbon ranks #2 of 52 on LMArena Agent's steerability signal (score 19.1), measuring how well the agent follows direction given mid-task.

    Best when: Its strongest result for this task is on LMArena Agent steerability, which measures following direction given mid-task.
    Source 12
    “Within the 52 catalog models matched to this evaluation: Ranks #2 of 52 on LMArena Agent's steerability signal (score 19.1), measuring how well the agent follows direction given mid-task.”
    LMArena Agent steerabilityOpen original ↗

    No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.

  8. 08

    GPT-6.1 TerraOpenAI · $1 in / $5 out per M tokens · 1.1M context

    GPT-6.1 Terra ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.

    Best when: Its strongest result for this task is on LMArena Agent tool hallucination, which measures calling only tools that exist, with real arguments.
    Source 13
    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.”
    LMArena Agent tool hallucinationOpen original ↗

    No practitioner reports are linked yet, so no cautions are listed. Absence of reports is not evidence of reliability.

Frequently asked

What is the best model for Tool & Function Calling?
Claude Fable 5.2 ranks first as of November 15, 2026. Ranks #1 on LMArena's agentic arena (score 30.4) and achieves perfect tool-hallucination and steerability scores, indicating strong multi-step tool-use performance with minimal fabricated calls.[1][2][3]
What is the best alternative to Claude Fable 5.2?
GPT-6 Astra ranks second. GPT-6 Astra ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.[4]

Sources

  1. 1

    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's steerability signal (score 33), measuring how well the agent follows direction given mid-task.”

    LMArena Agent steerability · Benchmark · Nov 11, 2026
  2. 2

    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.”

    LMArena Agent tool hallucination · Benchmark · Nov 11, 2026
  3. 3

    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena's agentic arena (score 30.4), measuring tool-use and multi-step task performance from human preference.”

    LMArena agentic arena · Benchmark · Nov 11, 2026
  4. 4

    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.”

    LMArena Agent tool hallucination · Benchmark · Nov 11, 2026
  5. 5

    “Within the 52 catalog models matched to this evaluation: Ranks #3 of 52 on LMArena Agent's steerability signal (score 18.6), measuring how well the agent follows direction given mid-task.”

    LMArena Agent steerability · Benchmark · Nov 11, 2026
  6. 6

    AI 新闻月刊(2026年10月)

    Original-language excerpt:“2026-10-29 清华姚期智团队提出 AWM:给陌生世界写定律并用 Lean 4 证明 - 清华交叉信息研究院姚期智、袁洋团队 10 月 29 日发布《Algebraic World Model》:把未知环境表示为「类型化对象 + 共享状态转换规则 + 目标条件」的显式数学模型,而非预测下一帧的神经网络。 - 流程三步:在环境中写可执行 Python 假设并用观测持续修正;用范畴论投射/归纳极限做结构化简;最后形式化为 Lean 4,证明目标可达或不可达。 - 两个「不可能通关」关卡:AWM 六次运行全部构造出内核检验通过的证明;配代码工具的 Claude Opus 5.5 与 GPT-6 Astra 在 12 次基线运行中一次都没做到。 - ARC-AGI-3 全部 25 个游戏、183 个关卡可让 Lean 复放通关轨迹;搭配 Opus 5 得分 99.9,获胜轨迹只用人类基线 43% 的动作。 - 相关链接:arXiv: [Algebraic World Model: From Exploration to P…”

    ikaijua · GitHub · Nov 2, 2026
  7. 7

    “Within the 15 catalog models matched to this evaluation: Ranks #8 of 15 on LMArena's search arena (score 1216), measuring answers produced with search.”

    LMArena search arena · Benchmark · Nov 9, 2026
  8. 8

    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.”

    LMArena Agent tool hallucination · Benchmark · Nov 11, 2026
  9. 9

    “Within the 15 catalog models matched to this evaluation: Ranks #5 of 15 on LMArena's search arena (score 1230), measuring answers produced with search.”

    LMArena search arena · Benchmark · Nov 9, 2026
  10. 10

    “The agent task benchmark (docs/PERFORMANCE.md → "Agent task benchmark") solved 24 of 24 runs on Claude Sonnet 5.5, so it can no longer show where the agent struggles or which change made it better. Record more per run (in tests/bench/agent_tasks.bench.ts): - model requests (#42); - tool calls by name (read_file, grep, edit_file, run_command, …), to see how a task was approached; - the reason for each failed call (see the issue on failed tool cards); - whether files changed without…”

    PierrunoYT · GitHub · Oct 1, 2026
  11. 11

    “Within the 52 catalog models matched to this evaluation: Ranks #4 of 52 on LMArena's agentic arena (score 16), measuring tool-use and multi-step task performance from human preference.”

    LMArena agentic arena · Benchmark · Nov 11, 2026
  12. 12

    “Within the 52 catalog models matched to this evaluation: Ranks #2 of 52 on LMArena Agent's steerability signal (score 19.1), measuring how well the agent follows direction given mid-task.”

    LMArena Agent steerability · Benchmark · Nov 11, 2026
  13. 13

    “Within the 52 catalog models matched to this evaluation: Ranks #1 of 52 on LMArena Agent's tool-hallucination signal (score 1), where a higher score means fewer fabricated tool calls.”

    LMArena Agent tool hallucination · Benchmark · Nov 11, 2026

Rankings synthesized from community evidence and open benchmarks. See methodology. Not driven by vendor marketing.