Latest
Google DeepMind launches Gemini 3.8 Flash and a cybersecurity version OpenAI says Astra is first model to cross Critical cybersecurity threshold
Telegram

Models

Google DeepMind launches Gemini 3.8 Flash and a cybersecurity version

The new model pushes deeper multi-step reasoning and tool use for coding agents, while the cyber-focused variant is limited to vetted defenders under a new access program called Fairwind.

  • Gemini 3.8 Flash costs $0.75 per million input tokens and $3.75 per million output tokens, matching 3.7 Flash's price.
  • Gemini 3.8 Flash scored 54.9% on the HLE-Verified reasoning benchmark, according to Google DeepMind.
  • Gemini 3.8 Flash Cyber found more than 70% of vulnerabilities in an internal test spanning 20 programming languages.

Who is ahead right now All leaderboards →

General questions, writing, reasoning

LMArena

Rank# Model Arena score
1 Claude Opus 5 Max Anthropic 1,505
2 Claude Opus 5 High Anthropic 1,504
3 Claude Opus 4.6 High Anthropic 1,503
4 Claude Opus 4.6 Anthropic 1,498
5 Claude Fable 5 Anthropic 1,494

Writing and fixing code

LMArena

Rank# Model Arena score
1 qwen3.8-max-0902 Alibaba 1,691
2 Claude Opus 5 Max Anthropic 1,688
3 Kimi K3 Max Moonshot AI 1,674
4 qwen3.8-max Alibaba 1,669
5 Claude Opus 5 High Anthropic 1,661

Cost per million tokens

OpenRouter

Rank# Model Per 1M tokens
1 Mistral Nemo Mistral $0.022
2 Ling-3.0-flash — $0.032
3 Granite 4.0 Micro IBM $0.041
4 Nex-N2-Mini Nex AGI $0.044
5 Solar Pro 4 Upstage $0.052

Updated 2026-09-02 · sources named on each board

By section

Models Releases, benchmarks, capability jumps

Policy Regulation, safety, copyright, courts

One line per story, the moment it lands.

Every article posted here goes to Telegram as a single sentence with the link. No digest, no inbox, nothing to unsubscribe from.

Open the channel