TokenPost topic
AI Models
44 TokenPost stories are tagged AI Models. The newest appear first.
Latest AI Models stories
- Arena's Agent Arena Lists Nearly 3 Million Sessions Across 65 Models
Claude Fable 5.2 (Max) led the Nov. 13 snapshot with a Net Improvement score of 28.61% ± 3.72%.
· Technology
- GPT-6.2 Astra Ranks Third as OpenAI Models Post Lower Task Costs
A Nov. 18 snapshot put Astra at 1,781 points and Sol at 1,757. Astra’s median task cost was about 3.8 times Sol’s.
· Technology
- Reson8’s Rumble-2-Flash API Debuted Nov. 3 at €0.01 per Audio Hour
The 500-million-parameter speech model transcribes prerecorded audio in more than 40 languages. Its repository requires users to accept access conditions before reaching the files.
· Technology
- 腾讯混元 Opens Three Hy4 flash Models Across Four Platforms
Hy4 flash, Hy4 flash-FP8 and Hy4 flash Base are available to download and deploy on GitHub, Hugging Face, ModelScope and GitCode.
· Technology
- Coinbase Proposes Marketplace for Specialized AI Models and Agents
Companies could offer customized models trained on proprietary data, while specialized agents could interact within the proposed marketplace.
· Technology
- Tencent’s Hy4 Flash Cuts Automated-Task Cost to 0.038 Yuan
The model scored 52.2 against Hy4’s 56.2 in the test. Tencent also reported lower costs on a PowerPoint collaboration benchmark.
· Technology
- Alibaba's Qwen3.8-Max Ranks 25th in Text, Third in Vision
The Nov. 11 text and Nov. 8 vision standings placed the model behind Claude Fable 5 in both categories.
· Technology
- Claude Fable 5.2 Costs Twice as Much as Opus 5.5 Per API Token
Both models have a 1-million-token context window. An offer giving eligible Pro and standard Team seats capped Fable access at no extra cost ends Nov. 17.
· Technology
- Baidu’s AI Marketplace Handles More Than 700 Billion Tokens a Day
A month after launch, 词元所 had more than 100 supply resources and averaged more than 50 million model calls a day.
· Technology
- OpenAI Proposes Nov. 12, 2026, for Wind-Down of Cursor Model Supply
The date is a proposal, and the official termination date awaits confirmation by both companies.
· Business
- MiniMax H3-Image License Bars Use in U.S., EU, U.K. and South Korea
The model’s weights are available on Hugging Face, while API images cost about $0.022 for 1K and $0.069 for 4K. Promotional discounts run through Tuesday, Nov. 17.
· Technology
- Codex Usage Statistics Include gpt-6.2-astra Identifier
A possible release could include Astra and Sol model lines and a cybersecurity-focused Cyber model; pricing and release channels remain unspecified.
· Technology
- Images Shifted Three AI Models’ Coding Choices in a Study
A red heart raised recursive solutions to 100% of samples for Claude Opus 4.7 and GPT-5.4; a yellow checkmark lifted Gemini 3.1 Pro to 90%.
· Technology
- Claude Fable 5.5 Catalog Entry Uses Specifications From Version 5.1
An Oct. 13 catalog update added the identifier `azure_ai/claude-fable-5-5`, but its price, context and capability fields came from Fable 5.1.
· Technology
- Claude Haiku 5.5 Matches GPT-6 Luna Rates for Prompts Up to 100,000 Tokens
Both models list rates of $0.10 per million input tokens and $0.50 per million output tokens in that tier. Anthropic charges more for longer prompts.
· Technology
- Codex Users Report `gpt-6.2-astra` in Usage Logs
One user-posted screenshot showed about 19.8% of requests assigned to the label, while other users said it did not appear in their records.
· Technology
- Liquid AI Debuts Six Open-Weight Models for Local AI Workloads
The family covers text, vision-language, audio and decision tasks, alongside an omni model that handles text, images, video and audio.
· Technology
- OpenAI Releases GPT-6.1 Terra With API Prices 50% Below Two Models
The model is available in ChatGPT Work and Codex. At medium reasoning effort, it scored 15 percentage points above GPT-5.6 Terra at about one-third the cost.
· Technology
- Anthropic Model Token Volume on OpenRouter Up 3.6x Since Opus 5.5 Launch
Claude Sonnet 5.5 and Haiku 5.5 followed by Oct. 7 with lower prices than Opus, but the platform data does not identify what drove the increase.
· Technology
- Box Evaluation Finds Claude Opus 5.5 Used About One-Third As Many Tokens As Opus 5
Box’s test found tasks finished 30% faster and deliverables were 42% shorter, with accuracy maintained. Opus 5.5 also has lower input and output prices than Opus 5.
· Technology
- Kimi K3.1 Pro Scores 64% on the Artificial Analysis Cyber Index
Kimi K3.1 Pro led publicly available models. Moonshot AI also detailed training changes aimed at reducing misuse risk before its planned full-weight release by Nov. 4.
· Technology
- Moonshot AI Releases Kimi K3.1 Pro, Plans Weight Release Before Nov. 4
A Highspeed version is also available, and both variants support context windows of up to 1 million tokens.
· Technology
- Venice Lists Mercury 3 for API Use With Prompts Visible to Providers
Venice's anonymized mode hides a user's identity from the model provider, but the provider can still see prompt content.
· Technology
- Google Cloud Adds OpenAI Models as Customers Gain Model-Switching Choice
Three OpenAI models are in private preview for select enterprise customers, with general availability planned in the coming weeks.
· Technology
- Top Chinese AI Models Trail U.S. Frontier by Estimated 6.5 Months
A separate software-task measure puts the lag at about 6.6 months. A March ranking found a 2.7% U.S. performance edge.
· Technology
- DeepSeek V4.1 Flash Tops Agent Arena Open-Source Rankings
The model ranked No. 11 overall in a result based on about 7,500 user sessions. Its median task cost was about $0.20.
· Technology
- Anthropic Prepares Model That May Infer Goals From Sparse Prompts
The examples are hypothetical, not published test results.
· Technology
- MiniMax Launches M3.1 Flash, Confirms Space Bunny Was Preview
The preview appeared as Space Bunny Alpha on OpenRouter on Sept. 23 and led OpenCode’s daily token-usage ranking by Sept. 24.
· Technology
- Perplexity Prices New Decision Models From $0.01 per Million Input Tokens
The 27B and 4B models handle text and images; output tokens are free, and the smaller model’s default format uses one question per request.
· Technology
- Claude Opus 5.5 Leads Agent Ranking as Gemini Tops Text List
Opus 5.5 ranked first in an Oct. 16 agent snapshot covering 2,534,501 sessions. Gemini 4 Argon led a separate text-preference ranking dated Oct. 15.
· Technology
- PrismML Packs Its 9B Bonsai 2 Model Into 2GB for Local AI
The 2B, 4B and 9B releases use main weights limited to −1, 0 and +1, with file sizes from 0.43GB to 2GB.
· Technology
- OpenRouter Users Sent 84% of Tokens to Non-Frontier Models in Aug. 10 Week
Six models handled about 80% of OpenRouter volume during the week of Aug. 10 while delivering roughly 77% of frontier-model performance at a fraction of the cost.
· Technology
- Meta Launches Muse Wonder AI Models as Shares Rise 3.38%
Muse Wonder and Muse Wonder Full are designed for complex tasks, including lease review and financial or health-information analysis.
· Technology
- 月之暗面 Released Kimi K3 in July
月之暗面 may be developing a 5 trillion- to 6 trillion-parameter model, while Kimi K3.1 may be released before mid-October.
· Technology
- Mistral Previews 1 Trillion-Parameter Large 4 AI Model
The open-weight system targets coding, cyberdefense, finance and other specialized work, with weights planned after safety testing.
· Technology
- Gensler Flags Chinese AI Models as Risk to U.S. Data-Center Returns
The former SEC chair said lower-cost, open-weight models could weaken U.S. pricing power as companies spend heavily on AI infrastructure.
· Technology
- AMD Agrees to Buy World Labs for $8.2 Billion in AI Push
The all-stock deal is expected to close by the end of 2026, subject to regulatory approvals. Fei-Fei Li will become AMD’s executive vice president and chief scientist.
· Business
- OpenAI and Anthropic Cut AI Model Prices as Competition Intensifies
OpenAI reduced GPT-5.6 Luna pricing by 80%, while Anthropic launched Claude Opus 5.5 at prices 20% below Opus 5.
· Technology
- Replit CEO Warns AI Application Companies Against Dependence on a Single Model Provider
Amjad Masad said Replit uses models from multiple providers and sometimes builds its own as pricing, performance and capabilities change.
· Technology
- Anthropic Plans Claude Sonnet 5.5 and Haiku 5.5 Launches
The company introduced Claude Opus 5.5 on Sept. 22 at $4 per million input tokens and $20 per million output tokens.
· Technology