TL;DR: DeepSeek's refreshed V4.1 Flash 1017 jumps 6 points to 45 on the Artificial Analysis Intelligence Index, tying for #2 among open-weights models at unchanged API pricing.
Summary: DeepSeek released DeepSeek V4.1 Flash 1017, an update to its flagship Flash model, which scores 45 on the Artificial Analysis Intelligence Index — up 6 points and just 1 behind MiMo-V2.6-Pro, tied with GLM-5.3 and ahead of Step 5 Preview and Kimi K3 at 44. The gains are concentrated in agentic work and reliability: Elo on GDPval-AA v2.1 rises 146 points to 1746, AA-Briefcase v1.1 climbs 150 points to 1570, and the AA-Omniscience non-hallucination rate jumps from 4% to 33% (accuracy 43%→46%), flipping that index from -5 to +5. Pricing on the first-party API is unchanged at $0.30 per 1M input and $1.20 per 1M output tokens at peak, with off-peak rates at half.
Why it matters: A same-price update that sharply cuts hallucination and boosts agentic benchmarks makes V4.1 Flash 1017 an easy drop-in upgrade for cost-sensitive agent pipelines. Re-run your eval suites against the 1017 snapshot, especially tool-use and factual-recall tasks, and consider scheduling batch work off-peak for half-price tokens.
Source: x_com