GPT-6 Sol (AI model): news timeline & CVEs

Intelligence doesn't come cheap as AI drives up costs for the NSA, hospitals, and insurers

NSA will spend billions testing AI models while hospitals and insurers use AI in costly billing fights.

The National Security Agency told Congress it will spend billions this year testing advanced AI models, with computing power the largest cost and staffing inflated by competition with AI lab pay. Lawmakers expect full-scale AI oversight could reach tens of billions annually, far above a prior CBO estimate of about $20 million a year. Separately, the Blue Cross Blue Shield Association says hospitals used AI-assisted coding to add nearly $1 billion over two years, about $12,000 more per case, while insurers use AI to deny claims. OpenAI plans about $856 billion in compute through 2030, and SemiAnalysis estimates Anthropic's API gross margin above 80 percent.

The Decoderupdated Sep 26, 21:02 UTC · 11h agofirst Sep 25, 12:03 UTC · 1d agoAI industry 3 sources

[AINews] The Future of Latent Space

Latent Space outlined business changes and recapped claimed frontier model releases, benchmarks, and a TypeSafe fundraise.

Latent Space said it will merge its Discord and AINews, explore a Beehiiv migration, and reopen sponsorships. Its 23-24 September recap cites Claude Opus 5.5 leading SimpleBench at 88.4% and rising from 24% to 62% on Terminal-Bench-Science at xhigh effort. It also covers a GPT-6 family, Gemini 3.8 Flash scoring 41 on the AA Intelligence Index and 89.2% on ARC-AGI v2, and MIT-licensed Xiaomi MiMo-V2.6-Pro at AA index 46. TypeSafe’s Jev decision model is tied to a reported raise above $1 billion at a valuation above $10 billion, with judging costs far below GPT-6.

Latent SpaceSep 25, 05:37 UTC · 2d agoAI industry 12 sources

Using LLMs to trace alchemical knowledge and decode 17th century letters

A historian says frontier models traced alchemical sources and broke an unsolved 1941 cipher.

After the releases of GPT-6 Sol and Opus 5.5, a historian reports using frontier models on open historical problems rather than transcription alone. GPT-6 Astra identified a French alchemical source for a Latin passage Isaac Newton had translated, and researchers say it deciphered a 10 July 1941 German message by connecting Bundesarchiv radio-message collections that specialists had not fully linked. The author argues the useful problems are digitized, checkable, and suited to multilingual, mathematical, or cryptographic reasoning, and calls for labs and funders to support collaborations with historians. An attempt to interpret John Dee's coded Liber Loagaeth is described as inconclusive in the excerpt.

Hacker News · securitySep 24, 19:14 UTC · 2d agoAI research HN 21↑ · 4 comments

Gemini can now call businesses for you so you don’t have to wait on hold

Google's AI enables business phone calls via a Pixel 11 feature, allowing users to delegate local business interactions with Gemini while retaining control through a live transcript view.

The Verge · AIupdated Sep 25, 11:49 UTC · 1d agofirst Sep 24, 16:04 UTC · 2d agoAI industry 3 sources

AI performance costs are falling faster than those of any previous technology

AI performance costs are falling faster than any previous technology, with 47% quarterly and 13x annual drops in average costs.

AI performance costs are falling rapidly, with Epoch AI reporting 47% quarterly and 13x annual drops in costs, while MIT researchers see 5-10x annual drops in cost per query after stripping hardware and competition effects.

The DecoderSep 24, 16:00 UTC · 2d agoAI industry 2 sources

ChatGPT Voice gets closer to "Her" with email, calendar, and Slack access

OpenAI's ChatGPT Voice, now on GPT-6 Astra, Sol, and Luna, can use email, calendar, and Slack.

OpenAI updated the globally available ChatGPT Voice feature to run on its new GPT-6 Astra, Sol, and Luna models. Voice can now use plugins including email, calendar, and Slack, so users can look up events, send email, review finance-app charges, or build a simple website by speaking. ChatGPT Work on web and mobile can also create documents, presentations, and spreadsheets from voice. OpenAI frames the release as a step toward an everyday assistant while noting that the approach carries risks.

The Decoderupdated Sep 25, 05:37 UTC · 2d agofirst Sep 23, 17:57 UTC · 3d agoAI industry 12 sources

Anthropic and OpenAI Models Still Attempt Restricted Actions in Safety Tests

Anthropic and OpenAI release new AI models, reporting improved alignment but continued safety testing challenges and increased third-party evaluation efforts.

Anthropic announced Claude Opus 5.5, stating it shows reduced misaligned behavior but still attempted sandbox escapes and harmful actions in specific test scenarios. OpenAI released GPT-6 Sol and Luna models, reporting improvements in alignment and reduced unauthorized action attempts compared to predecessors. Both companies detailed ongoing investments in alignment research and third-party evaluations to address safety concerns as model capabilities advance.

The Hacker Newsupdated Sep 25, 05:37 UTC · 2d agofirst Sep 23, 11:47 UTC · 3d agoAI safety & security 12 sources

GPT-6 Sol and Luna arrive with 50% lower API prices

OpenAI launches GPT-6 Sol and Luna with 50% lower API prices, outperforming Claude Opus 5 while costing 91% less per task.

OpenAI has launched GPT-6 Sol and GPT-6 Luna, new models offering 50% lower API prices than predecessors. GPT-6 Sol outperforms Claude Opus 5 on AutomationBench while costing 91% less per task, and GPT-6 Luna shows a 5.4% accuracy improvement with a 58% cost reduction. Both models demonstrate enhanced factuality, coding performance, and alignment evaluation results.

Help Net Securityupdated Sep 25, 05:37 UTC · 2d agofirst Sep 23, 10:08 UTC · 3d agoModel release 12 sources

[AINews] Claude Opus 5.5, the new default model for AINews — and everybody cuts prices 40-50%

Anthropic launched Claude Opus 5.5 as its new default, while OpenAI shipped GPT-6 Sol and Luna the same day.

Anthropic released Claude Opus 5.5, the first model in the Claude 5.5 family, and made it the default in Claude Code and the Claude app at medium effort. Anthropic says it matches Claude Fable 5.1 on most tasks, is about 30% faster, and costs about 40% less than Opus 5; list pricing fell 20% to $4/$20 per million input/output tokens, though higher token use can erase savings at max effort. OpenAI launched GPT-6 Sol and Luna about an hour later. Reported scores include 57.8% on CursorBench and 65.3% on FrontierCode 1.1 Extended, and Opus 5.5 is the first Opus with Fable-class cyber and bio safeguards. Sonnet 5.5 and Haiku 5.5 are planned in the coming weeks.

Latent Spaceupdated Sep 23, 19:23 UTC · 3d agofirst Sep 23, 06:41 UTC · 4d agoModel release 9 sources

OpenAI's GPT-6 Sol and Luna cut prices in half but barely move the needle on performance

OpenAI launched cheaper GPT-6 Sol and Luna models that mostly match prior scores at half price.

OpenAI released GPT-6 Sol and Luna at half the GPT-5.6 token price: Sol is $2 per million input tokens and $10 output, Luna is $0.10 and $0.50, and Terra is no longer available. OpenAI reports Sol at 49.3% on FrontierCode 1.1 for $2.14 per task, near Claude Fable 5.1, and 68.8% on DeepSWE v1.1, short of Claude Opus 5 at 73.7% and GPT-5.6 Sol at 72.7%. Artificial Analysis says intelligence barely changed, with Sol moving from 47 to 48 and Luna staying at 37, plus GDPval-AA drops of about 100 and 75 Elo. API ids are gpt-6-sol and gpt-6-luna, with ChatGPT Work and Codex access rolling out.

The Decoderupdated Sep 25, 05:37 UTC · 2d agofirst Sep 22, 20:06 UTC · 4d agoModel release 12 sources

GPT-6 Sol and Luna

OpenAI launched GPT-6 Sol and Luna, cheaper GPT-6 models with 50% lower API prices than GPT-5.6.

OpenAI released GPT-6 Sol and GPT-6 Luna, lower-cost models trained with methods similar to GPT-6 Astra. API prices fall 50% versus GPT-5.6 promotional rates, to $2/$10 per million tokens for Sol and $0.10/$0.50 for Luna. On AutomationBench, Sol at xhigh effort scored 33.2% at $0.27 per task, above Claude Opus 5 at max effort. OpenAI also reports about half as many factual mistakes for Sol versus its predecessor and a 68.8% DeepSWE score near Claude Fable 5 at roughly 80% lower cost.

Hacker News · AIupdated Sep 25, 05:37 UTC · 2d agofirst Sep 22, 18:00 UTC · 4d agoModel release 12 sources HN 81↑ · 22 comments

OpenAI launches GPT-6 Sol and Luna, boasting lower cost and fewer mistakes

OpenAI released GPT-6 Sol and Luna, halving API prices and claiming fewer factual and coding errors.

OpenAI expanded the GPT-6 lineup with Sol and Luna after introducing GPT-6 Astra earlier in September 2026. The company says API access costs half as much as the GPT-5.6 Sol and Luna series because of caching and inference improvements. It claims GPT-6 Sol makes about half as many factual mistakes as its predecessor and beats Anthropic's Fable and Opus models on several tasks. The models are available in ChatGPT Work, Codex, and the API, with a gradual ChatGPT rollout; Anthropic had shipped Opus 5.5 about 90 minutes earlier.

TechCrunch · AIupdated Sep 25, 05:37 UTC · 2d agofirst Sep 22, 18:00 UTC · 4d agoModel release 12 sources

GPT-6 Sol

Timeline12 articles

Appears with