PromptsZone Logo

PromptsZone

AI news & analysis
DaisyRose-27B Safetensors weight sizes: about 55.6 GB for BF16, 30.9 GB for FP8 and 21.9 GB for NVFP4. These sizes do not describe runtime memory.

DaisyRose-27B open weights: downloads and coding benchmarks

November 19th

Compare DaisyRose-27B Apache 2.0 downloads, the released 256K context window, and coding and instruction-following benchmarks against Qwen3.8-27B and Muse Glimmer.

Read more
New Models
Cloudflare’s self-hosted image-field test: Clairic returned its full response in 453 ms at p50 and 976 ms at p95; Clairic-flash took 305 ms and 605 ms.

Cloudflare opens Clairic and Clairic-flash weights for PDF extraction

November 19th

Compare Cloudflare Clairic and Clairic-flash on document extraction, hosted token pricing and latency, and find both Apache 2.0 model downloads.

Read more
New Models
Developer-run IFBench comparison: Olala mini 57.7%, Ling 3.0-tiny 57.3%, LFM2.5 8B-A1B 48.3%, Granite 4.0-H-Tiny 14.3%. Higher is better.

OVHcloud releases Olala mini 7A1B weights: downloads and benchmarks

November 19th

OVHcloud’s Olala mini 7A1B weights are out. Compare BF16 and GGUF downloads, context limits and five benchmarks against similarly sized models.

Read more
New Models
Microsoft-reported Aion Instruct benchmark scores for the unquantized model: IFEval 83.90%, GSM8K 74.91% and MMLU-Pro 33.35%.

Microsoft releases Aion Instruct weights ahead of Windows rollout

November 19th

Microsoft’s Aion Instruct 2.3B weights are available in bfloat16 and INT4 ONNX WebGPU. Compare the downloads, published scores and Windows rollout targets.

Read more
New Models
Exploration limits in Project Genie: Genie 4 supports about one minute of continuous interaction; Genie 3 supports up to three minutes per generation and can link worlds.

Genie 4 reaches Google AI Ultra: access, audio and session limits

November 19th

Check who can try Genie 4 in Project Genie, how live editing and native audio work, and when the updated Genie 3 is the better choice.

Read more
New Models
Qwen4-Max Model Studio API prices for Singapore: $1.50 standard input and $4.50 output, in US dollars per million tokens.

Qwen4-Max API pricing, availability and five benchmark comparisons

November 18th

See where Alibaba’s Qwen4-Max is available, what its Singapore API costs, and how it compares with Gemini 4 Neon and Claude Opus 5.5 on Vals benchmarks.

Read more
New Models
Vals Index v2.2: Gemini 4 Neon 70.9%, Claude Opus 5.5 70.1%, GPT-6.2 Sol 66.1%. Higher is better.

Gemini 4 Neon rolls out: pricing, access and six benchmark comparisons

November 18th

Gemini 4 Neon is rolling out on paid Gemini surfaces. Check API rates, model access and six benchmark comparisons with Claude Opus 5.5 and GPT-6.2 Sol.

Read more
New Models
Conceptual diagram comparing an agent report, runtime record and policy decision before expanding authority; not a released specification or product screen.

Microsoft launches AI Transparency Initiative: what agent evidence should show

November 18th

Microsoft's new AI Transparency Initiative targets independently verifiable agent behavior. Separate the standards effort from shipping products and useful audit evidence.

Read more
AI News
Selected voice benchmark comparison: MAI Realtime 1 leads Big Bench Audio at 98.9%; Gemini 3.8 Live Extended Thinking High leads the displayed τ-Voice scores at 68.6%. MAI figures are internal, peer figures attributed by Microsoft to Artificial Analysis.

MAI Realtime 1 enters preview: voice benchmarks, pricing and access

November 18th

Compare MAI Realtime 1 on two voice benchmarks, separate audio and reasoning charges, and check the limits of MAI Labs versus Foundry preview access.

Read more
New Models
Rollout diagram: free Tasks in Gemini in six countries; Google Pay, texting and email start in the U.S. with additional eligibility requirements.

Tasks in Gemini goes free in six countries: access and approval limits

November 18th

Google is expanding free Tasks in Gemini to six countries. Here is where payments, texting and email are rolling out, and what still needs approval.

Read more
AI News
Tencent-reported internal WorkIndex scores: Hy4 flash 42.1, Hy4 preview 39.7 and Hy4 51.4. Higher is better. The chart uses the official Tencent Hy family mark.

Hy4 flash weights are now available: downloads, benchmarks and license limits

November 18th

Tencent's Hy4 flash now has downloadable instruct, FP8 and base weights. Compare reported benchmarks, deployment requirements and the custom license.

Read more
New Models
Vercel AI Gateway output rates per million tokens: MAI Image 2.7 $34; MAI Image 2.7 Flash $10. Input charges are additional. Pricing, not a quality benchmark.

MAI Image 2.7 Flash and Canvas arrive: API access, prices and preview limits

November 18th

Microsoft adds MAI Image 2.7 Flash and Canvas. Compare API token rates, public-preview access and the separate personal-use limits of MAI Labs.

Read more
New Models
Microsoft-reported pass rates: MAI Code 2 versus Flash scores 73.2% versus 59.9% on DeepSWE 1.1, 84.6% versus 78.3% on TerminalBench 2.1, and 43.1% versus 44.8% on SWE Atlas. Higher is better; each row's higher value is bold.

MAI Code 2 and Flash arrive in Copilot: access, benchmarks and the migration deadline

November 17th

MAI Code 2 and Flash are available in GitHub Copilot. Compare three coding benchmarks, plan eligibility and the December 18 retirement of Code 1.1 Flash.

Read more
New Models
Microsoft-Decision-1.1 Foundry input pricing per million tokens: Global Standard $0.042, US Data Zone $0.046, EU and APAC Data Zones $0.05. Output tokens have no charge. Pricing tiers, not performance scores.

Microsoft-Decision-1.1 goes live: pricing and a practical routing check

November 17th

Microsoft-Decision-1.1 reaches general availability. Check Foundry pricing, Gateway migration details and how to test routing decisions before automating them.

Read more
New Models
GPT-6.2 Standard pricing per million tokens: Astra $8 input and $40 output; Sol $2 input and $10 output, for requests up to 272,000 input tokens. Price comparison, not performance.

GPT-6.2 Astra and Sol reach the API: pricing and a migration trap to check

November 17th

OpenAI adds GPT-6.2 Astra and Sol to the API. Compare token prices and context limits, and check the conversation changes that can discard reasoning state.

Read more
New Models
1/7
1 2 3 4 5 6 7

Newsletter Subscription

Receive the latest AI news

Blog logo

PromptsZone

AI news and analysis from PromptsZone: model launches, research, AI tools, and the business and policy developments shaping artificial intelligence.

RSS feed

© PromptsZone, 2026.

Made with Blogmaker

×