AI Pulse stories about LLMs & foundation models, newest first.
Anthropic released Claude Haiku 5.5 on Wednesday, the third model in its Claude 5.5 family in the past month, according to Reuters. Anthropic's news page describes it as its fastest, cheapest and most capable small model, aimed at high-volume, cost-sensitive work. Pricing and availability details were not confirmed in the pages reviewed.
Mistral chief executive Arthur Mensch told a conference in Abu Dhabi that the French company's latest model beats Chinese models in some areas, including cybersecurity. The claim is the company's own and not independently verified. It shows Mistral positioning itself as a European alternative to US and Chinese systems.
Google announced Gemini 4 Argon, its first new flagship since the Gemini 3 series. Access starts with trusted cyber defenders and US government pre-release testing, with paid API customers and AI Ultra subscribers to follow. VentureBeat reports it leads or ties on 13 of 18 disclosed benchmarks, though rivals still win on some coding tasks, with introductory API pricing of $2 per million input tokens and $10 per million output.
At its DevDay event, OpenAI revealed GPT-6.1 Sol, which Engadget's live coverage describes as replacing the shelved GPT-6.1 Astra. It also showed an Ultrafast mode said to generate about 300 tokens per second, plus 'Dots', always-on proactive agents offered to select users, and a shared ChatGPT workspace for teams. The launches arrive as OpenAI faces scrutiny over safety and deception issues in its models.
Anthropic released Claude Sonnet 5.5, the second model in its 5.5 family, positioning it as a faster, cheaper complement to Opus 5.5. It jumps to 70.6% on the Terminal-Bench agentic coding test from 10.3% for Sonnet 5, runs over 30% faster, and is the first Sonnet to ship with frontier-grade cyber safeguards. It's available now on Anthropic's platform, AWS, Google Cloud and Azure at unchanged pricing.
OpenAI released GPT-6 Sol and GPT-6 Luna on 22 September, minutes after Anthropic shipped Opus 5.5. Sol handles harder work like coding and costs $2 in and $10 out per million tokens. Luna is built for high-volume clerical jobs such as summarising and extraction, at $0.10 and $0.50. OpenAI says Sol makes about half as many factual mistakes as its predecessor. Its pitch is cost per completed task rather than raw benchmark scores, and for most business workloads that's the better measure.
UK angle: Luna's pricing makes bulk summarising and extraction cheap enough to stop worrying about. Test it against whatever you're paying for now.
Anthropic launched Claude Opus 5.5 on 22 September, the first model in its Claude 5.5 family. It says Opus 5.5 matches Fable 5.1 on most work and costs 40% less to run than Opus 5 on typical workloads, with API pricing cut to $4 per million input tokens and $20 per million output. Output is more than 30% faster too. It's the company's first release since Dario Amodei called for pacing the frontier, and plenty of people have pointed at the gap between the essay and the launch. Anthropic's answer is that METR and Frontier Design evaluated the model before release. Sonnet 5.5 and Haiku 5.5 follow within weeks.
UK angle: Teams building on Claude should redo their cost models. Opus-class work on typical workloads is now 40% cheaper.
That's everything. You've reached the first story.
