Best LLM for Summarizing Transcripts: From 30-Second Voicemails to Hour-Long Meetings

Choosing an LLM to summarize call transcripts depends heavily on transcript length, as short voicemails prioritize speed and price while long meetings require stronger coverage and faithfulness. Voice Standard Bench evaluates more than 50 LLM Gateway models on real speech-to-text transcripts at 30-second, 2-minute, 15-minute, and 60-minute lengths, using LLM judges to assess faithfulness, coverage, and conciseness rather than token-overlap metrics. As of October 28, 2026, gpt-5.6-terra leads 30-second summaries, Claude Opus 5.5 leads 2- and 15-minute calls, and DeepSeek V4.1 Flash has the highest 60-minute score, though its long-call results included notable request errors and high latency. Low-cost models such as Nemotron 3 Nano can be dramatically cheaper but generally score lower, while fast models may sacrifice quality, especially for longer calls. The benchmark expresses spending as cost per hour of audio to make voice-product budgeting clearer, recommends testing shortlisted models on an organization’s own calls, and advises using speaker-labeled transcripts, length-appropriate prompts, grounding instructions, and chunk-and-merge approaches for lengthy meetings.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.