List of Flash News about AI benchmark
| Time | Details |
|---|---|
|
2026-09-12 06:45 UTC |
Epoch AI: FrontierMath Benchmark Hits 98%
Epoch AI FrontierMath scores jumped from 5% to 98% in 14 months as GPT-6 Astra solved every Tier 4 problem, advancing AI solving previously unsolved mathematical problems research implications. |
|
2026-09-03 13:46 UTC |
OpenAI: GPT-6 Astra Saturates ARC-AGI-3
OpenAI GPT-6 Astra scores 63% on ARC-AGI-3 with 99% adapter harness, topping humans on 96% of tasks and saturating the AGI benchmark. |
|
2026-07-31 16:00 UTC |
OpenAI: Two API Settings Triple ARC-AGI-3 Scores
OpenAI: Two API settings tripled GPT-5.6 scores on ARC-AGI-3 benchmark while boosting efficiency through retained reasoning and compaction. |
|
2025-12-30 16:04 UTC |
GPT-5.2 Pro Near FrontierMath Tier 4 Benchmark: Catalyst Watch for AI Traders
According to @gdb, GPT-5.2 Pro is very strong for science and mathematics, and the post notes that reaching FrontierMath Tier 4 would evidence the complex reasoning needed for scientific breakthroughs, with the model described as getting very close. Source: twitter.com/gdb/status/2006154439208337417. The Tier 4 description cited in the post comes from the FrontierMath site, which states that solving Tier 4 would provide evidence an AI can perform the complex reasoning required for breakthroughs in technical domains. Source: FrontierMath official site as referenced in twitter.com/gdb/status/2006154439208337417. No benchmark scores, release timelines, or model-card details are provided in the post, so no formal performance verification is available from the source. Source: twitter.com/gdb/status/2006154439208337417. No cryptocurrencies or tokens are mentioned in the post; a confirmed market catalyst would require formal benchmark results or leaderboard updates verifying a Tier 4 solution. Source: twitter.com/gdb/status/2006154439208337417 and FrontierMath official site. For trading relevance, the only verifiable signal today is the capability claim itself; confirmation risk remains until benchmark authorities publish results. Source: twitter.com/gdb/status/2006154439208337417 and FrontierMath official site. |
|
2025-12-29 17:07 UTC |
Gensyn (GENS) Delphi Benchmark Final Eval 11/11 Goes Live — Market Paused Dec 29–Jan 7, Full Results Posted
According to @gensynai, Eval 11 of 11 for the Gensyn Middleweight General Reasoning Benchmark market on Delphi is live, marking completion of the evaluation cycle (source: @gensynai on X, Dec 30, 2025). According to @gensynai, this is the final benchmark for this specific Delphi market track, confirming no further evals are scheduled in this run (source: @gensynai on X, Dec 30, 2025). According to @gensynai, no market will be running from Dec 29 through Jan 7, indicating a scheduled pause on Delphi for this benchmark window (source: @gensynai on X, Dec 30, 2025). According to @gensynai, traders can review the complete benchmarking results at https://github.com/gensyn-ai/delphi-middleweight-reasoning to inform timing and analysis specific to this Delphi benchmark track (source: @gensynai on X, Dec 30, 2025). According to @gensynai, the concluded eval and defined pause window set the near-term catalyst timeline for participants tracking Gensyn (GENS) and AI compute benchmarks on Delphi for this market only (source: @gensynai on X, Dec 30, 2025). |
|
2025-12-27 12:41 UTC |
Gensyn launches Eval 10 of 11 for Middleweight General Reasoning Benchmark on Delphi with full results on GitHub (2025 AI benchmark update)
According to @gensynai, Eval 10 of 11 of the Gensyn Middleweight General Reasoning Benchmark market on Delphi is now live (source: https://twitter.com/gensynai/status/2005016298993189175). The full benchmarking results are publicly available in the official repository at https://github.com/gensyn-ai/delphi-middleweight-reasoning, enabling market participants to directly review the published outcomes and methodology from the primary source (source: https://github.com/gensyn-ai/delphi-middleweight-reasoning). |
|
2025-12-25 12:26 UTC |
Gensyn Releases Eval 9 Results for Delphi Middleweight General Reasoning Benchmark — Full AI Metrics Now Live on GitHub
According to @gensynai, Eval 9 of 11 for the Gensyn Middleweight General Reasoning Benchmark market on Delphi is live, and the full benchmarking results are available on Gensyn’s GitHub repository, providing traders with immediate access to verifiable performance data. Source: Gensyn on X, Dec 25, 2025; Gensyn GitHub |
|
2025-12-23 12:57 UTC |
GPT-5.2 Exceeds Human Baseline on ARC-AGI-2: AI Benchmark Milestone and Trading Takeaways
According to @gdb, GPT-5.2 exceeded the human baseline on the ARC-AGI-2 benchmark in a post on December 23, 2025, signaling a notable AI capability milestone, source: https://twitter.com/gdb/status/2003570781192957991. The announcement did not disclose numerical scores, methodology, or evaluation settings, limiting immediate comparability and independent verification, source: https://twitter.com/gdb/status/2003570781192957991. The post includes no release timing, commercial details, or any references to cryptocurrencies, tickers, or tokens, which constrains direct trading implications until further disclosures, source: https://twitter.com/gdb/status/2003570781192957991. |
|
2025-12-16 11:36 UTC |
Greg Brockman Unveils Expert-Level AI Scientific Reasoning Benchmark; 2026 Called a Year of Acceleration — AI Crypto Tokens RNDR, FET, AGIX in Focus
According to @gdb, 2026 will be a year of scientific acceleration through AI, and he announced a new benchmark to measure AI capability in expert-level scientific reasoning (source: Greg Brockman on X, Dec 16, 2025). AI milestone announcements have previously coincided with notable rallies in AI-linked crypto assets; Reuters reported sharp gains in FET, RNDR and AGIX during the early 2023 ChatGPT-driven frenzy, underscoring the sensitivity of these tokens to AI news flow (source: Reuters, Feb 2023). Traders tracking the AI narrative can reference this new benchmark as a catalyst when building watchlists and risk scenarios for AI-focused tokens (source: Greg Brockman on X, Dec 16, 2025; Reuters, Feb 2023). |
|
2025-12-16 09:25 UTC |
OpenAI Launches FrontierScience Benchmark for PhD-Level Reasoning Across Physics, Chemistry, Biology — Trading Takeaways for AI Stocks and Crypto
According to @sama, OpenAI released FrontierScience, a new evaluation to measure expert-level scientific reasoning with PhD-level difficulty across physics, chemistry, and biology, featuring expert-written olympiad-style problems and longer research-style tasks designed to reveal where models succeed and where they fall short (Source: @sama on X; OpenAI announcement on X, Dec 16, 2025). For traders, the post describes an evaluation release only and does not include model scores, system-to-system comparisons, API changes, product launch details, or any mention of cryptocurrencies or tokens, limiting immediate, quantifiable catalysts based on the disclosed information (Source: @sama on X; OpenAI announcement on X, Dec 16, 2025). |
|
2025-12-16 09:04 UTC |
OpenAI Launches FrontierScience AI Benchmark for PhD-Level Scientific Reasoning Across Physics, Chemistry, Biology: Key Takeaways for AI Stocks and Tokens
According to @OpenAI (source: @OpenAI on X, Dec 16, 2025), the company released FrontierScience, a new evaluation benchmark to measure expert-level scientific reasoning at the PhD level. According to @OpenAI (source: @OpenAI on X, Dec 16, 2025), FrontierScience spans physics, chemistry, and biology and uses hard, expert-written questions, including olympiad-style problems and longer formats. According to @OpenAI (source: @OpenAI on X, Dec 16, 2025), the post does not include benchmark scores, model rankings, or performance comparisons, limiting immediate quantitative signals for AI-focused equities and AI-related tokens. According to @OpenAI (source: @OpenAI on X, Dec 16, 2025), the announcement does not mention crypto, blockchain, tokens, pricing, or API changes, and it presents FrontierScience as an evaluation resource rather than a commercial product release, indicating no specified direct catalyst for crypto markets in the post. |
|
2025-12-11 10:18 UTC |
OpenAI: GPT-5.2 Thinking Hits Human Expert Level on GDPval Across 44 Occupations — What Traders Should Know
According to @OpenAI, GPT-5.2 Thinking is its first model to reach human expert-level performance on GDPval, a benchmark covering well-specified knowledge-work tasks across 44 occupations, including making presentations and spreadsheets; source: OpenAI on X, Dec 11, 2025. The announcement cites GDPval and task types but does not disclose score breakdowns, methodology details, release timing, or deployment information; source: OpenAI on X, Dec 11, 2025. For traders, this is a capability milestone headline with no direct crypto or market data provided in the source, making it a narrative update rather than a quantified trading signal; source: OpenAI on X, Dec 11, 2025. |