Speko LLM Benchmark for Voice Agents v3
Models use simulated support tools to look up orders and issue refunds. The task score gives partial credit when a model stops early without causing harm.
Speko reports task scores for 30 model configurations. Each runs 19 scenarios five times. Separate probes check whether models invent policies.
- Made by
- Speko
- Kind
- Customer service
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Agent task
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools, A simulator
- Refresh
- New versions replace it
- Run by
- A host
- Based on
- A version of another benchmark
- Strongest model tested
- GPT-5.6 Terra #18 on the capability index
- Last model added
- November 2026
- Scores refreshed
- November 20th, 2026
Gemini 3.5 Flash-Lite leads with 94.7% Gemini 3.5 Flash-Lite leads with 94.7%
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
- 94.7
Gemini 3.5 Flash-Lite
- 93.2
DeepSeek V4 Flash 0731
- 92.6 GPT-4.1 mini
- 91.8 GPT-5.4 Mini
- 90.9 Grok 4.3
- 89.1 GPT-4.1
- 86.8
Nemotron 3 Ultra
- 86.8
Gemini 3.7 Flash
- 86.8
Gemini 4 Hydrogen
- 86.1
Qwen 3.8 27B
- 85.8
Gemini 3.8 Flash
- 85.3
Gemma 4 31B IT
- 84.2
DeepSeek V4.1 Flash
- 83.5 GPT-5.6 Terra
- 83.3 Claude Sonnet 5
- 82.5 Claude Haiku 5.5
- 82.0 gpt-oss-120b
- 81.9 GPT-5 mini
- 81.7 Claude Haiku 4.5
- 80.9 GLM-4.7
- 80.5 Mercury Voice
- 77.0 GPT-5 nano
- 75.6 GPT-5
- 75.6 GPT-5.2
- 75.6 GPT-5.6 Luna
- 54.3
Llama 3.3 70B
- 30.6
Qwen-Turbo
- 1
Gemini 3.5 Flash-Lite Gemini 94.7%
- 2
DeepSeek V4 Flash 0731 Baseten 93.2%
- 3 GPT-4.1 mini OpenAI 92.6%
- 4 GPT-5.4 Mini OpenAI 91.8%
- 5 Grok 4.3 xAI 90.9%
- 6 GPT-4.1 OpenAI 89.1%
- 7
Nemotron 3 Ultra Baseten 86.8%
- 8
Gemini 3.7 Flash Gemini 86.8%
- 9
Gemini 4 Hydrogen Gemini 86.8%
- 10
Qwen 3.8 27B Cerebras 86.1%
- 11
Gemini 3.8 Flash Gemini 85.8%
- 12
Gemma 4 31B IT Hopper 85.3%
- 13
DeepSeek V4.1 Flash Together 84.2%
- 14 GPT-5.6 Terra OpenAI 83.5%
- 15 Claude Sonnet 5 Anthropic 83.3%
- 16 Claude Haiku 5.5 Anthropic 82.5%
- 17 gpt-oss-120b Cerebras 82%
- 18 GPT-5 mini OpenAI 81.9%
- 19 Claude Haiku 4.5 Anthropic 81.7%
- 20 GLM-4.7 Baseten 80.9%
- 21 Mercury Voice Inception 80.5%
- 22 GPT-5 nano OpenAI 77%
- 23 GPT-5 OpenAI 75.6%
- 24 GPT-5.2 OpenAI 75.6%
- 25 GPT-5.6 Luna OpenAI 75.6%
- 26
Llama 3.3 70B Together 54.3%
- 27
Qwen-Turbo Alibaba 30.6%
Source: Official Speko LLM Benchmark for Voice Agents v3 leaderboard
- 94.7
Gemini 3.5 Flash-Lite
- 93.2
DeepSeek V4 Flash 0731
- 92.6 GPT-4.1 mini
- 91.8 GPT-5.4 Mini
- 90.9 Grok 4.3
- 89.1 GPT-4.1
- 86.8
Nemotron 3 Ultra
- 86.8
Gemini 3.7 Flash
- 86.8
Gemini 4 Hydrogen
- 86.1
Qwen 3.8 27B
- 85.8
Gemini 3.8 Flash
- 85.3
Gemma 4 31B IT
- 84.2
DeepSeek V4.1 Flash
- 83.5 GPT-5.6 Terra
- 83.3 Claude Sonnet 5
- 82.5 Claude Haiku 5.5
- 82.0 gpt-oss-120b
- 81.9 GPT-5 mini
- 81.7 Claude Haiku 4.5
- 80.9 GLM-4.7
- 77.0 GPT-5 nano
- 75.6 GPT-5
- 75.6 GPT-5.2
- 75.6 GPT-5.6 Luna
- 54.3
Llama 3.3 70B
- 30.6
Qwen-Turbo
- 1
Gemini 3.5 Flash-Lite Gemini 94.7%
- 2
DeepSeek V4 Flash 0731 Baseten 93.2%
- 3 GPT-4.1 mini OpenAI 92.6%
- 4 GPT-5.4 Mini OpenAI 91.8%
- 5 Grok 4.3 xAI 90.9%
- 6 GPT-4.1 OpenAI 89.1%
- 7
Nemotron 3 Ultra Baseten 86.8%
- 8
Gemini 3.7 Flash Gemini 86.8%
- 9
Gemini 4 Hydrogen Gemini 86.8%
- 10
Qwen 3.8 27B Cerebras 86.1%
- 11
Gemini 3.8 Flash Gemini 85.8%
- 12
Gemma 4 31B IT Hopper 85.3%
- 13
DeepSeek V4.1 Flash Together 84.2%
- 14 GPT-5.6 Terra OpenAI 83.5%
- 15 Claude Sonnet 5 Anthropic 83.3%
- 16 Claude Haiku 5.5 Anthropic 82.5%
- 17 gpt-oss-120b Cerebras 82%
- 18 GPT-5 mini OpenAI 81.9%
- 19 Claude Haiku 4.5 Anthropic 81.7%
- 20 GLM-4.7 Baseten 80.9%
- 21 GPT-5 nano OpenAI 77%
- 22 GPT-5 OpenAI 75.6%
- 23 GPT-5.2 OpenAI 75.6%
- 24 GPT-5.6 Luna OpenAI 75.6%
- 25
Llama 3.3 70B Together 54.3%
- 26
Qwen-Turbo Alibaba 30.6%
Source: Official Speko LLM Benchmark for Voice Agents v3 leaderboard
| # | Entry | Score ± | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | | 94.7% 79 to 100 | Gemini | Jul 2026 | Official leaderboard | |
| 2 | | 93.2% 83 to 100 | Baseten | Jul 2026 | Oct 2026 | Official leaderboard |
| 3 | GPT-4.1 mini | 92.6% 80 to 100 | OpenAI | Apr 2025 | Official leaderboard | |
| 4 | GPT-5.4 Mini | 91.8% 84 to 98 | OpenAI | Mar 2026 | Oct 2026 | Official leaderboard |
| 5 | Grok 4.3 | 90.9% 78 to 100 | xAI | Apr 2026 | Official leaderboard | |
| 6 | GPT-4.1 | 89.1% 75 to 100 | OpenAI | Apr 2025 | Official leaderboard | |
| 7 | | 86.8% 74 to 97 | Baseten | Jun 2026 | Oct 2026 | Official leaderboard |
| 8 | | 86.8% 71 to 100 | Gemini | Aug 2026 | Oct 2026 | Official leaderboard |
| 9 | | 86.8% 71 to 100 | Gemini | Nov 2026 | Nov 2026 | Official leaderboard |
| 10 | | 86.1% 72 to 97 | Cerebras | Aug 2026 | Official leaderboard | |
| 11 | | 85.8% 71 to 98 | Gemini | Sep 2026 | Oct 2026 | Official leaderboard |
| 12 | | 85.3% 68 to 100 | Hopper | Apr 2026 | Oct 2026 | Official leaderboard |
| 13 | | 84.2% 69 to 97 | Together | Sep 2026 | Oct 2026 | Official leaderboard |
| 14 | GPT-5.6 Terra | 83.5% 68 to 96 | OpenAI | Jul 2026 | Oct 2026 | Official leaderboard |
| 15 | Claude Sonnet 5 | 83.3% 68 to 95 | Anthropic | Jun 2026 | Official leaderboard | |
| 16 | Claude Haiku 5.5 | 82.5% 65 to 96 | Anthropic | Oct 2026 | Oct 2026 | Official leaderboard |
| 17 | gpt-oss-120b | 82% 70 to 93 | Cerebras | Aug 2025 | Official leaderboard | |
| 18 | GPT-5 mini | 81.9% 69 to 94 | OpenAI | Aug 2025 | Official leaderboard | |
| 19 | Claude Haiku 4.5 | 81.7% 65 to 94 | Anthropic | Oct 2025 | Official leaderboard | |
| 20 | GLM-4.7 | 80.9% 67 to 92 | Baseten | Dec 2025 | Oct 2026 | Official leaderboard |
| 21 | Mercury Voice | 80.5% 68 to 91 | Inception | Oct 2026 | Official leaderboard | |
| 22 | GPT-5 nano | 77% 63 to 90 | OpenAI | Aug 2025 | Official leaderboard | |
| 23 | GPT-5 | 75.6% 58 to 91 | OpenAI | Aug 2025 | Official leaderboard | |
| 24 | GPT-5.2 | 75.6% 59 to 91 | OpenAI | Dec 2025 | Oct 2026 | Official leaderboard |
| 25 | GPT-5.6 Luna | 75.6% 58 to 91 | OpenAI | Jul 2026 | Oct 2026 | Official leaderboard |
| 26 | | 54.3% ±15.5 | Together | Dec 2024 | Official leaderboard | |
| 27 | | 30.6% 12 to 52 | Alibaba | Nov 2024 | Official leaderboard |
| # | Model | Score ± | Setting | Released | Run on | Source |
|---|---|---|---|---|---|---|
| 1 | | 94.7% 79 to 100 | Gemini | Jul 2026 | Official leaderboard | |
| 2 | | 93.2% 83 to 100 | Baseten | Jul 2026 | Oct 2026 | Official leaderboard |
| 3 | GPT-4.1 mini | 92.6% 80 to 100 | OpenAI | Apr 2025 | Official leaderboard | |
| 4 | GPT-5.4 Mini | 91.8% 84 to 98 | OpenAI | Mar 2026 | Oct 2026 | Official leaderboard |
| 5 | Grok 4.3 | 90.9% 78 to 100 | xAI | Apr 2026 | Official leaderboard | |
| 6 | GPT-4.1 | 89.1% 75 to 100 | OpenAI | Apr 2025 | Official leaderboard | |
| 7 | | 86.8% 74 to 97 | Baseten | Jun 2026 | Oct 2026 | Official leaderboard |
| 8 | | 86.8% 71 to 100 | Gemini | Aug 2026 | Oct 2026 | Official leaderboard |
| 9 | | 86.8% 71 to 100 | Gemini | Nov 2026 | Nov 2026 | Official leaderboard |
| 10 | | 86.1% 72 to 97 | Cerebras | Aug 2026 | Official leaderboard | |
| 11 | | 85.8% 71 to 98 | Gemini | Sep 2026 | Oct 2026 | Official leaderboard |
| 12 | | 85.3% 68 to 100 | Hopper | Apr 2026 | Oct 2026 | Official leaderboard |
| 13 | | 84.2% 69 to 97 | Together | Sep 2026 | Oct 2026 | Official leaderboard |
| 14 | GPT-5.6 Terra | 83.5% 68 to 96 | OpenAI | Jul 2026 | Oct 2026 | Official leaderboard |
| 15 | Claude Sonnet 5 | 83.3% 68 to 95 | Anthropic | Jun 2026 | Official leaderboard | |
| 16 | Claude Haiku 5.5 | 82.5% 65 to 96 | Anthropic | Oct 2026 | Oct 2026 | Official leaderboard |
| 17 | gpt-oss-120b | 82% 70 to 93 | Cerebras | Aug 2025 | Official leaderboard | |
| 18 | GPT-5 mini | 81.9% 69 to 94 | OpenAI | Aug 2025 | Official leaderboard | |
| 19 | Claude Haiku 4.5 | 81.7% 65 to 94 | Anthropic | Oct 2025 | Official leaderboard | |
| 20 | GLM-4.7 | 80.9% 67 to 92 | Baseten | Dec 2025 | Oct 2026 | Official leaderboard |
| 21 | GPT-5 nano | 77% 63 to 90 | OpenAI | Aug 2025 | Official leaderboard | |
| 22 | GPT-5 | 75.6% 58 to 91 | OpenAI | Aug 2025 | Official leaderboard | |
| 23 | GPT-5.2 | 75.6% 59 to 91 | OpenAI | Dec 2025 | Oct 2026 | Official leaderboard |
| 24 | GPT-5.6 Luna | 75.6% 58 to 91 | OpenAI | Jul 2026 | Oct 2026 | Official leaderboard |
| 25 | | 54.3% ±15.5 | Together | Dec 2024 | Official leaderboard | |
| 26 | | 30.6% 12 to 52 | Alibaba | Nov 2024 | Official leaderboard |
4 models have held number one on the official leaderboard since 2023
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Speko LLM Benchmark for Voice Agents v3 leaderboard
GPT-4.1 mini on the official leaderboard passed 85% in April 2025
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Speko LLM Benchmark for Voice Agents v3 leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Tool use
- AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state.
Gemini 4 Neon 56.3%
- MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks.
Gemini 4 Argon 90.7%
- BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes.
Qwen3.7-Max 75%
- τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state.
Qwen 3.8 Max 55.1%
- Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
- DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Task access check. We don’t know whether the tasks are public. Every benchmark
Catalog record 1 warning 1 pass
- Warning Task access We don’t know whether the tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (27 entries).
- Pass Models scored 26 models we track, plus 1 other entry on its leaderboard.
- Pass Newest score The newest is from November 2026, for Gemini 4 Hydrogen.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is GPT-5.6 Terra (#18).
- Note Headroom Saturated. The best is 94.7%, past 85% since April 2025.
Leaderboard import 3 pass
- Pass Importer We read the board on speko.com: 29 rows for models we track and 1 other entry at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.