Recommendation for Tool calling
Tool & Function Calling
Our top recommendation for Tool & Function Calling, based on the public evidence we track, is Anthropic: Claude Opus 4.5.[1][2] Use for complex multi-turn tool chains where hallucination resistance and format fidelity matter, as it ranks #1 on BFCL's comprehensive suite. Watch out: Expect ~4.38s mean latency per BFCL measurement, which may bottleneck real-time interactive workflows. Anthropic: Claude Sonnet 4.5 is the next-ranked alternative. Enable strict: true (constrained decoding) in Bedrock Converse when receiving structured tool outputs to prevent PageDecodeError from type drift.
About this recommendation
- Updated
- Sep 25, 2026
- Evidence through
- Sep 25, 2026
- Sources
- 17
- Revision
- v80
Decision audit
Why this result
Inspect the inputs and the computed order behind the recommendation.
Models screened
20
live candidates
Evaluation feeds
5
task-weighted
Winner coverage
64%
intended feed weight
Largest provider share
4 of 8
Anthropic
Sources evaluated
The task sets these weights before any model is scored.
| Evaluation feed | Weight | Winner result | Field measured |
|---|---|---|---|
| Berkeley Function Calling | 55% | #1 | 6/20 |
| LMArena Agent | 15% | not measured | 12/20 |
| LMArena Search | 10% | not measured | 4/20 |
| OpenRouter usage | 10% | 65/100 | 20/20 |
| Structured-output evalunavailable | 10% | feed unavailable | 0/20 |
Provider concentration
Each exact model is scored separately; provider identity is not a ranking input.
- Anthropic4 models
- Z.ai2 models
- Google1 model
- OpenAI1 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.
| Rank | Model | Relative score | Coverage | Practitioner evidence | Strongest measured reason |
|---|---|---|---|---|---|
| 01 | Claude Opus 4.5Anthropic | 67 | 64% | 1 threads · 1 families · 0 cautions | #1 Berkeley Function Calling |
| 02 | Claude Sonnet 4.5Anthropic | 66 | 64% | 2 threads · 2 families · 2 cautions | #2 Berkeley Function Calling |
| 03 | GLM 4.6Z.ai | 63 | 64% | 1 threads · 1 families · 0 cautions | #3 Berkeley Function Calling |
| 04 | Claude Haiku 4.5Anthropic | 63 | 64% | 2 threads · 2 families · 1 cautions | #4 Berkeley Function Calling |
| 05 | GPT-5.6 SolOpenAI | 63 | 36% | 22 threads · 15 families · 12 cautions | #1 LMArena Search · #4 LMArena Agent |
| 06 | GLM 5.2Z.ai | 57 | 34% | 13 threads · 8 families · 5 cautions | #11 LMArena Agent |
| 07 | Claude Fable 5Anthropic | 54 | 36% | 4 threads · 4 families · 0 cautions | #2 LMArena Agent · #3 LMArena Search |
| 08 | Gemini 2.5 FlashGoogle | 53 | 64% | 3 threads · 3 families · 2 cautions | #8 Berkeley Function Calling |
Relative score combines normalized benchmark quality and signal coverage; independent practitioner evidence and freshness are bounded tie-breakers. It is an ordering score, not an absolute quality percentage. The writing model receives this order and cannot change it.
Leads BFCL tool-calling benchmark with 77.47% overall, covering single-turn, multi-turn, web-search, memory, hallucination, and format-sensitive scenarios.
Best when: Use for complex multi-turn tool chains where hallucination resistance and format fidelity matter, as it ranks #1 on BFCL's comprehensive suite.
Tips
- Use for complex multi-turn tool chains where hallucination resistance and format fidelity matter, as it ranks #1 on BFCL's comprehensive suite.
Watch out for
- Expect ~4.38s mean latency per BFCL measurement, which may bottleneck real-time interactive workflows.
Requires strict schema enforcement via
strict: trueto guarantee type fidelity in tool outputs, as non-strict mode permits type mismatches and missing required fields.Best when: Enable `strict: true` (constrained decoding) in Bedrock Converse when receiving structured tool outputs to prevent `PageDecodeError` from type drift.
Tips
- Enable `strict: true` (constrained decoding) in Bedrock Converse when receiving structured tool outputs to prevent `PageDecodeError` from type drift.
Watch out for
- Image editing workflows fail even with native function calling due to invalid URL construction in tool call arguments.
Ranks #3 on BFCL with 72.38% overall, competitive on multi-turn and format-sensitive tool calling despite chat template quirks in the GLM family.
Best when: Deploy for cost-sensitive tool-calling pipelines where BFCL #3 performance at 4.34s latency trades off against open-weight flexibility.
Tips
- Deploy for cost-sensitive tool-calling pipelines where BFCL #3 performance at 4.34s latency trades off against open-weight flexibility.
Watch out for
- Chat template lacks explicit thinking on/off switches, relying on `reasoning_effort` and `clear_thinking` knobs that may interact unpredictably with reasoning extraction.
Benchmarks at roughly half the Intelligence Index of GLM-5.2 (24 vs 51), with observed 422 validation errors in GitLab Duo integration.
Best when: Consider only after reviewing the cited caution.
Watch out for
- GitLab Duo returns 422 validation errors on initial tool calls, suggesting provider-specific integration fragility.
Ranks #1 on LMArena search arena (1257) and leads implementation work in agent routing protocols, though tool-effort overrides can bypass configured subagent limits.
Best when: Assign to implementation-heavy agent tasks where search-integrated answers and protocol leadership matter, per Sparq routing adoption.
Tips
- Assign to implementation-heavy agent tasks where search-integrated answers and protocol leadership matter, per Sparq routing adoption.
Watch out for
- Tool `effort: "hi"` overrides `task.agentModelOverrides`, forcing `max` tier even when subagent is explicitly configured to `medium` or `high`.
- Exhibits transport errors and fetch failures in long-running sessions, with 19 of 37 transport-error events in one study.
Ranks #10 on LMArena agentic arena with practical GUI automation deployments, though prone to tool-call spam loops and reasoning/command split failures.
Best when: Integrate into local-first agent loops with 14 native tools for GUI automation, as demonstrated in the Nolan project with voice-driven task execution.
Tips
- Integrate into local-first agent loops with 14 native tools for GUI automation, as demonstrated in the Nolan project with voice-driven task execution.
Watch out for
- Tends to spam 100+ redundant tool calls in production agentic systems compared to more restrained sequencing in Gemini models.
- Enters failure loops where correct bash commands are planned in `thinking` but emitted tool calls contain truncated or malformed leading segments like `cd && `.
Positioned for plan-review and advisory roles in model-tier taxonomies, though forced tool choice support was removed in upstream pydantic_ai for this generation.
Best when: Reserve for security review and code advisory tasks where ultra-tier reasoning is preferred over execution, per advisor-board recommendations.
Tips
- Reserve for security review and code advisory tasks where ultra-tier reasoning is preferred over execution, per advisor-board recommendations.
Watch out for
- Forced tool choice (`tool_choice='required'`) is unsupported; upstream flipped rejection families to exclude this model from constrained tool selection.
Scores 56.24% on BFCL with natural tool sequencing in real scenarios, though model is deprecated and exhibits silent provider fallback issues.
Best when: Prefer over spam-prone alternatives for intuitive multi-tool sequences in production agentic systems, per practitioner reports.
Tips
- Prefer over spam-prone alternatives for intuitive multi-tool sequences in production agentic systems, per practitioner reports.
Watch out for
- Model is deprecated with imminent discontinuation; also exhibits silent fallback to wrong model version when configured candidate 404s.
Frequently asked
- What is the top-ranked model for Tool & Function Calling?
- Anthropic: Claude Opus 4.5 ranks first in the current evidence-weighted comparison. Use for complex multi-turn tool chains where hallucination resistance and format fidelity matter, as it ranks #1 on BFCL's comprehensive suite.[1]
- What should I watch out for with Anthropic: Claude Opus 4.5?
- Expect ~4.38s mean latency per BFCL measurement, which may bottleneck real-time interactive workflows.[1]
- What is an alternative to Anthropic: Claude Opus 4.5?
- Anthropic: Claude Sonnet 4.5 is the next-ranked option. Enable `strict: true` (constrained decoding) in Bedrock Converse when receiving structured tool outputs to prevent `PageDecodeError` from type drift.[2]
Sources
- 1
“Scores 77.47% overall on BFCL (#1 of 31), measuring single-turn, multi-turn, web-search, memory, hallucination, and format-sensitive tool calling; mean latency was 4.38s.”
Berkeley Function-Calling Leaderboard · Benchmark · Sep 24, 2026 - 2
“## 背景 #111 で経路 B の応答を tool use で受け取るようにしたが、2026-08-22 23:09 の再投入 (2608.17950v1) で 12 ページ目がモデルの出力 `sections` が配列ではなく文字列で返り、`PageDecodeError` になった (#113 の再試行でも同じ)。 tool use の `inputSchema` は `strict` を付けないと型を保証しない。公式ドキュメントは「strict なしでは型の不一致や必須欠落が起こり得る」と述べ、解決策として `strict: true` (constrained decoding) を提示している。 Claude Sonnet 4.5 は Bedrock Converse (`bedrock-runtime`、cross-region inference 含む) で対象。 - [ToolSpecification API Reference](https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime…”
tamaco489 · GitHub · Aug 22, 2026 - 3
“I retested with `anthropic/claude-sonnet-4.5` using Native function calling. The behavior is different from GPT-4.1 Mini, but image editing still does not work. The image upload itself succeeds: However, I observed two different failures. ### Failure 1: Open WebUI builds an invalid image URL edit_image attempts to load the uploaded image from: This returns: The same file is available successfully through the local Open WebUI endpoint: The relevant traceback is: ### Manual attachment fallback wi…”
GUILHERME-GARCIATECH · GitHub · Jul 28, 2026 - 4
“Scores 72.38% overall on BFCL (#3 of 31), measuring single-turn, multi-turn, web-search, memory, hallucination, and format-sensitive tool calling; mean latency was 4.34s.”
Berkeley Function-Calling Leaderboard · Benchmark · Sep 24, 2026 - 5
“### 🐛 Describe the bug GLM-5.3's chat template has **no thinking on/off switch** — its only knobs are `reasoning_effort` (`'low'`/`'high'`, anything else → `'max'`) and `clear_thinking`: (`zai-org/GLM-5.3-Flash` `chat_template.jinja`; `enable_thinking` appears nowhere in the template.) But the GLM reasoning parser still gates reasoning **extraction** on the GLM-4.5/4.6-era kwargs. At current `main`, `vllm/parser/glm47_moe.py`: So when a client sends `chat_template_kwargs: {"enable_thinking": f…”
sethforprivacy · GitHub · Sep 1, 2026 - 6
“### OmniRoute Version 3.8.49 ### Installation Method npm (global) ### Operating System Linux ### OS Version Ubuntu 22.04.5 LTS (kernel 5.15.0-1106-nvidia) ### Node.js Version 24.18.0 ### Provider(s) Involved GitLab Duo ### Model(s) Involved gitlab-duo/claude-sonnet-4-6, gitlab-duo/claude-haiku-4-5, gitlab-duo/claude-sonnet-4-6-high ### Client Tool Claude Code v2.1.220 (launched via omniroute launch --profile) ### Description GitLab Duo returns 422 {"detail":"Validation error"} on the very first…”
jayparmar88 · GitHub · Jul 31, 2026 - 7
“> 🤖 SPARQ agent\n\nAdopt the maintainer's new default model protocol in Sparq's target routing catalog:\n\n- GPT-5.6 Sol leads implementation work.\n- Claude Opus 5 owns review and soundness.\n- Preserve the restricted security/soundness posture and explicit exhaustion behavior.\n- Keep target PLAN routing and registry CLAIM routing in exact agreement.\n\nUpdate the routing table and its regression tests; do not modify protected agent briefs.”
jeswr · GitHub · Sep 2, 2026 - 8
“Ranks #1 of 6 on LMArena's search arena (score 1257), measuring answers produced with search.”
LMArena search arena · Benchmark · Aug 24, 2026 - 9
“### Description Since 17.1.2, the `task` tool accepts `effort: "lo" | "med" | "hi"`. The tool prompt tells the calling model to choose one based on task complexity, so models routinely send `"hi"` even for relatively simple tasks. `"hi"` maps to the highest level supported by the model and takes precedence over `task.agentModelOverrides`. With `gpt-5.6-sol`, this means `max`, even when I explicitly configure the subagent as `medium` or `high`. There is no setting to cap or disable this override…”
mixmav · GitHub · Jul 25, 2026 - 10
“## 背景 Pi Web Desktop 中 openai-codex（ChatGPT 订阅 OAuth）会话频繁出现「输出一半停止 / fetch failed / terminated」，其他模型（aliyun/deepseek）偶发但次数少。 ## 实锤数据（2026-08-01，scripts/session-stops.mjs 修复后全量扫描 185 会话） - transport-error 共 37 条：openai-codex/GPT 19（gpt-5.6-sol×11 + gpt-5.6-terra×8）、aliyun 9、pi-router 8、deepseek 1。 - 最大异常会话：311 消息 / ~2.2M tokens，会话开头连续 6 次 fetch failed/terminated。 - compaction 仅 1 次且在非中断点，**不是主因**；主因是 transport error。 ## 工具修复（commit bfe020d） session-stops.mjs 原判定只在独立 type=error 事件时识别 transport-er…”
dust617 · GitHub · Aug 1, 2026 - 11
“## 项目地址 https://github.com/xcq20100224/Nolan ## 一句话介绍 市面上的 AI 助手只会给建议，Nolan 直接动手：对它说「打开网易云音乐，播放我喜欢列表里的第一首歌」，它会真的移动鼠标去完成。 ## 主要特点 - **真·动手**：语音指令直接驱动 GUI 自动化——打开软件、点击、播放音乐、写文件、定时叫醒，不是输出聊天建议 - **本地优先**：语音识别用本地 faster-whisper，隐私不出本机；语音合成 GLM-TTS 带离线兜底 - **架构轻**：GLM-5.2 Agent 循环 + 14 个本地工具；React 前端 + Python 标准库后端，无 Electron - **一键安装**：install.bat 自动探测 Python 环境，配一个免费智谱 API key 即可运行 ## 演示 真机演示 GIF（自动打开网易云放歌全过程）： https://raw.githubusercontent.com/xcq20100224/Nolan/main/docs/demo.gif README 里还附了一次 Rea…”
xcq20100224 · GitHub · Jul 31, 2026 - 12
“some of the tool calling is better, its better at knowing how to use a sequence of tools in a real world scenario. things like glm 5.2 will spam tool calls like 100 times. gemini model will just use the tools as you would expect im always convinced people with takes on the open source models have never actually used them in a production agentic system”
llmslave · Hacker News · Jul 10, 2026 - 13
“## Environment - **Model:** glm-5.2 (GLM Coding Plan, Anthropic-compatible endpoint) - **Harness:** Claude Code CLI v2.1.215 (`ANTHROPIC_BASE_URL` pointed at Z.ai) - **Context:** agentic coding session, Japanese-language context, ~450 assistant turns at the time of failure - **Date observed:** 2026-07-21 (JST) ## Summary During error recovery, GLM-5.2 entered a loop where it **planned the correct Bash command verbatim inside `thinking`, then emitted a tool call with the leading `cd && ` segment…”
carrotRakko · GitHub · Jul 21, 2026 - 14
“## Summary A five-model advisor review of the newly merged model-tier taxonomy (`c266410`, agent-harness#309) converged on an author/reviewer split, but rejected making the heavy tier the default implementation tier. Recommended upstream policy: - roadmap/plan/supervise: Claude Opus 5 (`heavy`) - execute/repair: Claude Sonnet 5 (`regular`), with a recorded escalation to Opus 5 after repeated verification failure - plan-review/code-review/advise/security: Claude Fable 5 (`ultra`) - worker/econom…”
ViperJuice · GitHub · Jul 26, 2026 - 15
“Ports upstream `pydantic_ai` changes (`3a3e5612786c64e19312f20e3c998553edf1353b..716f2ae4a1cb2650ce4ee58f702d1f60c3c93f8c`) into the Go port. ## Ported - `ai/models/anthropic/anthropic.go` (`anthropicSupportsForcedToolChoice`): upstream flipped the rejected families for forced tool choice from `claude-fable-5`, `claude-mythos-5`, `claude-mythos-preview` to `claude-fable-5-1`, `claude-mythos-5-1`. Anthropic's per-model forcing-tool-use table now lists only the 5.1 generation, and `claude-fable-5…”
github-actions[bot] · GitHub · Sep 8, 2026 - 16
“Might be a model behavior issue or upstream provider issue for not wrapping the thought but honesty, i am guessing this is a model behavior issue here. Also gemini 2.5 is being deprecated any day now and is a dinosaur. Shouldn't rely on it for any form of tool calling really. I wouldn't at least. Not an open webui issue. Not reproducible with any newer gemini model or other models.”
Classic298 · GitHub · Aug 10, 2026 - 17
“Independent confirmation on Linux x64, omp/18.3.0 (the reporter's env was macOS/18.2.10): the bug is still present after 18.2.10. - `omp q "current date"` with `modelRoles.web: google-antigravity/gemini-3.8-flash:high` silently serves `Provider: gemini-2.5-flash @ Gemini` — the 404'd configured candidate is dropped with no visible error (the #6868/#5300 visibility gap). - `omp q --model google-antigravity/gemini-3.8-flash "hi"` → `Gemini Cloud Code API error (404): Requested entity was not foun…”
bannert1337 · GitHub · Sep 24, 2026
Rankings synthesized from community evidence and open benchmarks. See methodology. Not driven by vendor marketing.