· 3 min read
The Gradient — 2026-10-18
- Industry News
- Company Blog
Replit has announced a strategic shift, naming OpenAI its primary AI model provider while still offering Anthropic's Claude as an alternative.
Replit makes OpenAI its default over Anthropic, CEO says
What happened: Over the weekend, Replit CEO Amjad Masad confirmed that OpenAI will replace Anthropic as the platform’s default model provider. Existing users can still access Claude, but new projects will default to OpenAI’s models.
Why it matters: The move signals Replit’s confidence in OpenAI’s performance and ecosystem, potentially streamlining developer workflows and attracting more AI‑centric users. Keeping Claude available preserves choice for developers who prefer Anthropic’s offerings.
Key stats:
- OpenAI becomes the default model for all new Replit projects.
- Claude remains accessible as an opt‑in option.
- No immediate pricing changes were announced.
Source: https://techcrunch.com/2026/10/17/replit-is-no-longer-working-with-anthropic-ceo-says/
Bringing GPT-Rosalind to academic research labs
What happened: OpenAI announced the Rosalind Academic Plan, a program that provides qualified academic research labs with free, sponsored access to the GPT‑Rosalind language model and a suite of tools tailored for life‑sciences experiments.
Why it matters: By lowering cost barriers and offering cutting‑edge AI capabilities, the plan aims to speed up hypothesis generation, data analysis, and experimental design across universities, potentially leading to faster breakthroughs in biology, genetics, and drug discovery.
Key stats:
- Eligible labs receive full‑access credits covering compute and API usage.
- The program opens to a limited cohort of research groups each year, with a focus on early‑career investigators.
- OpenAI estimates that AI‑augmented labs could cut research cycle times by up to 30%.
Source: https://openai.com/index/introducing-the-rosalind-academic-plan
Elorian’s first AI model wants to reason with images, not just about them
What happened: Elorian released Glimpse, an AI model designed to reason directly with images. In internal tests, Glimpse surpassed the newer GPT‑6 Astra on scientific chart reasoning and short‑form video captioning, while delivering competitive scores on a range of other visual‑reasoning tasks.
Why it matters: By moving from image description to true visual reasoning, Glimpse opens doors for AI to interpret data visualizations, videos, and complex imagery—key capabilities for research, education, and media automation.
Key stats: • Glimpse outperformed GPT‑6 Astra on scientific chart reasoning benchmarks. • It also led on short‑form video captioning tests. • Performance remained competitive across additional visual reasoning datasets.
Source: https://techcrunch.com/2026/10/16/elorians-first-ai-model-wants-to-reason-with-images-not-just-about-them/
Amp puts AI agents to work beyond code with OpenAI models
What happened: Amp integrated OpenAI’s GPT‑6 Astra, Ultrafast, GPT‑Live‑1, and a “Sign in with ChatGPT” feature into its platform, enabling agents to perform tasks outside traditional programming environments.
Why it matters: This pushes AI agents into everyday applications, demonstrates the scalability of large language models, and lowers the barrier for developers to embed AI in non‑coding contexts.
Key stats: • GPT‑6 Astra claims roughly 2× faster inference than its predecessor. • Ultrafast cuts latency to sub‑50 ms for real‑time queries. • GPT‑Live‑1 supports continuous streaming responses. • Early adopters report up to a 30% boost in productivity for non‑coding workflows.
How Third Moment Research runs 10,000 agents for 36+ hours
What happened:
Third Moment Research launched a fleet of 10,000 autonomous AI agents that remained active for 36+ hours straight. The agents were powered by the latest GPT‑6 Astra and GPT‑6.1 Sol models, with specialized tools like Codex for code generation and Daybreak for cybersecurity tasks.
Why it matters:
Sustained, large‑scale agent deployment proves that advanced language models can maintain focus over days, adapt through continuous use, and handle diverse workloads—from software development to threat detection—without human intervention. This marks a step toward truly persistent AI assistants.
Key stats:
- 10,000 agents running concurrently
- 36+ hours of uninterrupted operation
- Models used: GPT‑6 Astra, GPT‑6.1 Sol, Codex, Daybreak
- Applications spanned coding, security monitoring, and autonomous decision‑making