What shipped |
● OpenAI · GPT · released Sep 3 · added this week OpenAI's new flagship: far fewer made-up answers and a third of the tokens on coding-agent work, but the same overall score as GPT-5.6 Sol — at 2.5 times the price. Compared with GPT-5.6 Sol · the model it replaces 51% ▼ down from 92% · 41 points fewer Invents facts on half as many trick questions · AA hallucination rate | $10 / $50 up from $4 / $20 · 2.5× the price Costs 2.5 times more to run than Sol · Price per 1M tokens |
|
● Anthropic · Claude Fable · released Sep 1 · added this week Anthropic's top model, refreshed: big jumps on long coding and scientific-research tasks, and cached input now costs 75% less — same headline price as Fable 5. Compared with Claude Fable 5 · the model it replaces 55.8% ▲ up from 42.0% · +14 points Finishes over half of long multi-step coding tasks · Terminal-Bench 4.0 | $10 / $50 same as Fable 5 · cache reads now $0.25 Same price — and cached input now costs 75% less · Price per 1M tokens |
|
● Google · Gemini Flash · released Sep 2 · added this week Google's new Flash: a big jump on command-line tasks and a few points smarter, at the same price — though it burns about 30% more tokens per task, and the price doubles in 2027. Compared with Gemini 3.7 Flash · the model it replaces 90.8% ▲ up from 81.6% · +9 points Finishes 9 in 10 multi-step tasks at the command line · Terminal-Bench 2.1 | $0.75 / $3.75 unchanged · doubles to $1.50 / $7.50 on Jan 1, 2027 Same price as 3.7 Flash — until the end of 2026 · Price per 1M tokens |
|
Also new this week |
● OpenAI · GPT Image · image model · Sep 8 Ships as two variants in ChatGPT and the API: Flare, with higher quality than GPT Image 2 at 50% lower latency, and Sunburst, a slower model for precise editing work. |
● Meta · Muse · released Sep 2 · added this week Meta's flagship edges closer to the leaders: four points up on the all-round score and clear gains on office and terminal tasks, at the same price and the same 1M-token context. |
● Google · Gemini Omni · video model · released Aug 27 · added this week Scene extension now reads 10 seconds of prior footage (up from 1) for clips up to 40 seconds, adds first-and-last-frame camera control, 4K upscaling and a 360p draft mode that renders up to 60% faster at a third of the cost. |
● DeepSeek · DeepSeek Flash · open weights · Sep 10 A new encoder-decoder design that activates only 8B parameters on input and 16B on output, beating V4-Flash on every agentic test DeepSeek published while keeping the MIT licence and 1M-token context — though the API price more than doubles. |
● Microsoft · MAI Image · image model · released Sep 4 · added this week No. 2 on the Arena text-to-image leaderboard, +79 Elo over MAI-Image-2.5 across all eight categories. |
● Alibaba · Wan · video model · open weights · released Aug 24 · added this week Doubles the maximum clip length to 30 seconds in a single pass, accepts documents (PDF, PPT, DOC) as input alongside text, image, audio and video, and improves face realism and reference consistency. |
● Alibaba · Qwen Flash · open weights · released Aug 26 · added this week Open-weight mixture-of-experts model with 125B total parameters but only 6B active per token, a 262K native context window and a built-in vision encoder, previewing the Qwen4 architecture. |
|
Next AI Model checks 9 AI makers for new models every morning. Every number comes from the maker or an independent test, with its source. One email a week, only when something ships. |
