#Benchmarks — DailyPrompt

▮#Benchmarks

AI9 Jul 2026 · 1 min

Insilico Medicine's AI-designed drug clears Phase IIa trial

ISM001-055 is the first drug both designed by AI and targeting an AI-discovered disease target to post positive mid-stage results.

Models7 Jul 2026 · 1 min

Google delays Gemini 3.5 Pro to July 17 for a full architecture rebuild

Google is scrapping the existing 2.5 Pro architecture after enterprise testers flagged excessive token consumption in extended agentic tasks.

Models7 Jul 2026 · 1 min

Gemini 3.5 Pro to ship with 2M-token context and a Deep Think layer

The rebuilt model targets mathematical reasoning, SVG scene generation and image quality, alongside a doubled context window and autonomous workflow capabilities.

Models7 Jul 2026 · 2 min

Anthropic releases Claude Fable 5, its first Mythos-class model

Fable 5 tops agentic coding leaderboards and introduces a new safety tier system that separates general availability from approved-organization access. Demo content.

Tools7 Jul 2026 · 1 min

Kling 3.0 tops the AI video leaderboard as Sora exits

Chinese models now dominate the text-to-video arena, with Kling, Seedance and Happy Horse leading a category OpenAI is stepping back from.

Models7 Jul 2026 · 1 min

xAI's V9 foundation model completes training, Grok 4.5 next

The ground-up architecture redesign finished training in late May and will underpin Grok 4.5, with xAI planning monthly V9-based model releases through 2026.

Tools6 Jul 2026 · 1 min

Ideogram's new style codes claim 78% photorealism in blind tests

With 4.3 billion accessible style presets and 90-95% text-rendering accuracy, Ideogram is closing the realism gap with professional photography.

Models6 Jul 2026 · 2 min

Google ships Gemini 3.5 Flash with 10M token context at commodity prices

The new Flash tier reads entire codebases and multi-hour video in a single request, and Google is pricing it aggressively against GPT-5-mini. Demo content.

Open Source6 Jul 2026 · 1 min

Open source models close the gap: DeepSeek V4, Qwen 3.7 Max, Llama 4

Open-weight models now trail frontier proprietary systems by single-digit percentages on everyday tasks, while costing four to ten times less per token.

Open Source25 Jun 2026 · 2 min

Local models catch up: 30B open weights now match last year's frontier

Benchmark analysis shows the gap between open 30B models and closed frontier systems has compressed to roughly twelve months, and consumer hardware runs them. Demo content.