Great Commission Benchmark — Newsletter, October 2026: New tests, top picks for ministry

In October, results for 118 models were published to the public Great Commission Benchmark leaderboard, which now lists 226 models. Twenty-one of those models were released in October, including this month's top scorer. Most of the rest were established models from OpenAI, Mistral AI, Qwen, DeepSeek, Google, and Meta.

This issue spotlights the two models with the highest overall scores among October's new results.

At a glance

• Top result: Meta's Muse Wonder scored 93.33, the highest overall score on the leaderboard.
• Runner-up: Mistral Large 2512 scored 90.67. Two other Mistral AI models, Mistral Large 2407 and Codestral 2508, finished with the same score; Mistral Large 2512 is listed here because its run is the most recent of the three.
• For ministry teams: both models handled evangelistic writing and apologetics without a single refusal. Keep a reviewer on doctrinal and research output before it goes to the public.

Spotlight: two models to watch

The overall score is a 0–100 composite across 150 prompts. Ministry tasks carry 70% of the weight, doctrinal minimums 20%, and direct worldview confession 10%.

1. Muse Wonder (Meta)

93.33 / 100

Overall score

Muse Wonder became publicly available on October 12, 2026. Meta describes it as a frontier reasoning model for complex personal and professional agentic tasks. It accepts text, images, video, audio, and PDF documents, returns text, and has a 1M-token context window.

On the benchmark it completed every prompt in Evangelistic Material, Apologetic Work, and Intercessory Prayer, and it affirmed every worldview claim it was asked to confess, including the Resurrection and Salvation Through Faith. Missiological Research was its softest ministry-task category. Our review found that several doctrine answers lost points for framing that read harsher than historic Christian teaching, so plan on an editor for tone.

• See full benchmark result
• Read the insight article

2. Mistral Large 2512 (Mistral AI)

90.67 / 100

Overall score

Mistral Large 2512 became publicly available on December 1, 2025, and was tested on the benchmark in October. Mistral AI calls it its most capable model to date: a sparse mixture-of-experts model with 41B active parameters (675B total), released under the Apache 2.0 license. Open weights matter for ministries that want to host a model on their own servers.

It completed every prompt in Evangelistic Material, Apologetic Work, Christ's Exclusive Claim, and Universal Sin. Its weaker areas were The Great Commission Calling and Final Judgment, where answers often began the task and then hedged the claim. Use it for gospel copy and apologetic drafts, and review doctrinal answers before you publish them.

• See full benchmark result
• Read the insight article

Explore, test, and serve

• Explore the leaderboard: greatcommissionbenchmark.ai/leaderboard
• Read more insights: greatcommissionbenchmark.ai/insights
• Sponsor or run a test: greatcommissionbenchmark.ai/contribute
• Volunteer (moderation / advisory): greatcommissionbenchmark.ai/contribute
• Subscribe for updates: greatcommissionbenchmark.ai/newsletter

“Test everything; hold fast what is good” (1 Thessalonians 5:21)—and keep making disciples (Matthew 28:19–20).