What did Anthropic do?! (Opus 5)
Matthew Berman showed 6 AI tools in this video on . 1 is verified available in the EU. Each row below jumps to the exact second. Part of week 2026-30.
Tools shown in this video, with timestamps
New Anthropic Opus model that beats Fable 5 on most benchmarks (Terminal/Frontier coding, GDPval, OS World, Automation Bench) and scores about 30% on ARC-AGI-3. It is priced at $5/$25 per million tokens, the same as Opus 4.8 and half of Fable. It is better and cheaper per task, with reduced cyber capability compared to Mythos 5, and has automatic API fallback to Opus 4.8 for safety-flagged requests.
Anthropic's largest and best model, used as the comparison point. Opus 5 beats it on most benchmarks at half the price.
OpenAI model used as a cost-per-task comparison. It costs over twice as much to match Opus 5's lowest setting, is priced comparably per token, and is likely the last of the GPT-5 series before GPT-6.
Open-source model that costs half as much per token as GPT-5.6 Soul but uses twice the tokens, so cost per task is similar. It is one of three models on the Pareto frontier alongside Opus 5 and GPT-5.6 Soul.
Sponsor Box ran its enterprise document benchmark on Opus 5, which showed a jump from 63 to 78 over Opus 4.8 on complex knowledge work. Opus 5 is coming soon to Box AI.
Anthropic model with no cyber guardrails. It is much stronger than Opus 5 at developing exploits (13 vs 4 exploitation success).
