x.com/dubrovnik07/status/2103543413933064530
1 correction found
9 frontier systems
Raycaster’s own benchmark materials describe a 12-system leaderboard, not 9 systems.
Full reasoning
Raycaster’s official Biopharma Bench announcement, published September 22, 2026 (three days before this post), says the benchmark reports results across "all twelve frontier systems" and shows a leaderboard for 12 frontier models. Raycaster’s benchmark results page likewise says "Benchmark results for 12 evaluated models" and lists 12 systems individually.
So the post’s claim that the benchmark covered "9 frontier systems" does not match the benchmark’s own published materials. Other parts of the post (71 assignments / 12 companies / GPT-6 Astra with 8 full passes) do appear in Raycaster’s materials, but the system count is stated as 12, not 9.
2 sources
- Biopharma Bench: Can Frontier Agents Do Real Professional Work? | Raycaster
September 22, 2026 ... Figure 1 reports the headline leaderboard across all twelve frontier systems ... Benchmark results across 12 frontier models.
- Biopharma Bench V0.1 - regulatory and CMC AI agent benchmark
Benchmark results for 12 evaluated models ... Rank ... 1 GPT-6 Astra ... 12 GPT-5.6 Sol ...