New AI models, what they cost, how they did in our tests, and how to use them. RSS
· News, Anthropic
Anthropic released Claude Fable 5.2 on 9 November 2026 at $8 / $40 per million tokens, down from $10 / $50. What changed, the benchmark claims, and the catch on cost per task.
· News, Meta
Meta's smallest and cheapest Muse model, released 9 November 2026, is built for the routine parts of agent work: tool calls, routing, extraction and coding subtasks. Price, limits and how it compares.
· News, Open weights, Moonshot AI
Moonshot AI published the weights of Kimi K3.1 Pro on Hugging Face in early November 2026. It ranks first among open-weight models on Arena's agent leaderboard. Size, licence and what it takes to run.
· News, Open weights, Mistral
Mistral published the weights of Mistral Large 4 on 29 October 2026: about 1 trillion parameters, roughly 50 billion active, 1M context and image input. The licence, the API price and the hardware it needs.
· Image models, New on SparrowCodes
We added six image models to SparrowCodes, from $0.011 an image. Same prompt for each, with the results, the time each took and what it costs per image.
· Coding models, New on SparrowCodes
Six coding models from $0.035 per million input tokens, now on SparrowCodes. We gave them the same ten programming tasks and an agent repair job. Here's how each did and what it cost.
· Sparrow Auto, Coding models
Sparrow Auto picks a model for each coding task: a low-cost one for routine work, a stronger one when the task is hard or failing. In our test it passed everything Sonnet passed at 8% of the cost.
