{"assertions":[{"assertion_index":0,"assertion_text":"On DeepSWE, a software-engineering test, Claude Fable 5 solves 69.7% of tasks on the first try against 62.8% for DeepSeek's new V4 Pro 0813 Together AI Blog.","citation_check":null,"claim_ids":[48029,48030,48031],"link_only_urls":[]},{"assertion_index":1,"assertion_text":"Z.ai chose to hold its new GLM-5.3 close at launch because the model got better at hacking-related work: access started with select partners and its own coding plan, with an open-weight release to follow a safety review Latent Space.","citation_check":null,"claim_ids":[48016,48017,48018],"link_only_urls":[]},{"assertion_index":2,"assertion_text":"Z.ai says API access and downloadable weights on Hugging Face are scheduled about two weeks after launch, and that the model came from extra post-training, meaning additional tuning after the main training run, on the same 743-billion-parameter base as GLM-5.2, parameters being the internal settings that mark a model's size Interconnects.","citation_check":null,"claim_ids":[46910,46913],"link_only_urls":[]},{"assertion_index":3,"assertion_text":"The safety nonprofit SaferAI had already placed GLM-5.2 just a few months behind GPT-5.5 and Claude Opus 4.7 on cyber and bio capability TechCrunch AI.","citation_check":null,"claim_ids":[42344,42353],"link_only_urls":[]},{"assertion_index":4,"assertion_text":"Anthropic's August risk report discloses an internal-only model, which the report labels Model 2, that substitutes for Anthropic's own researchers on 62.8% of a measured slice of their work, up from 54.8% for the earlier Mythos Preview Don't Worry About the Vase.","citation_check":null,"claim_ids":[48595,48596,48597],"link_only_urls":[]},{"assertion_index":5,"assertion_text":"In February the company argued that catastrophic risk from its models automating research was very low, a position METR reviewed at the time METR.","citation_check":null,"claim_ids":[21826,21827],"link_only_urls":[]},{"assertion_index":6,"assertion_text":"The rules took effect July 15, and ByteDance's Doubao and Alibaba's Tongyi Qianwen shut down user-created human-like companions, ending attachments some users experienced as losing a partner ChinaAI.","citation_check":null,"claim_ids":[48139,48140,48141],"link_only_urls":[]},{"assertion_index":7,"assertion_text":"That complicates the earlier line that the US restrains its frontier AI more than China restrains its own frontier models Don't Worry About the Vase.","citation_check":null,"claim_ids":[7316],"link_only_urls":[]},{"assertion_index":8,"assertion_text":"Researchers probing Ai2's fully open Olmo 3 found that for 51% to 59% of drugs tested, the model showed little sign of holding any drug-specific knowledge, and another 12% to 18% of answers tracked the drug name's suffix rather than stored facts Allen Institute for AI.","citation_check":null,"claim_ids":[48364,48365,48366],"link_only_urls":[]},{"assertion_index":9,"assertion_text":"- Z.ai's safety review of GLM-5.3 will show whether gated-first releases become standard practice for Chinese labs or a one-off. Latent Space","citation_check":null,"claim_ids":[48016,48017,48018],"link_only_urls":[]},{"assertion_index":10,"assertion_text":"- Anthropic's next risk report is the checkpoint for whether the researcher-substitution number keeps climbing while Model 2 stays internal. Don't Worry About the Vase","citation_check":null,"claim_ids":[48595,48596,48597],"link_only_urls":[]},{"assertion_index":11,"assertion_text":"- OpenAI says Asana cleared five years of engineering work in two weeks with Codex; buyer-side replications outside vendor pages are what would make that number spendable. OpenAI News","citation_check":null,"claim_ids":[48361,48362],"link_only_urls":[]},{"assertion_index":12,"assertion_text":"- Alibaba's Qwen 3.8 27B, an Apache-licensed vision model small enough for a laptop, is drawing praise alongside complaints that it overthinks by default. Simon Willison's Weblog","citation_check":null,"claim_ids":[47586],"link_only_urls":[]},{"assertion_index":13,"assertion_text":"- Mojo, the GPU-focused programming language, finally shipped as open source under an Apache 2 license after a promise first made in 2023. Simon Willison's Weblog","citation_check":null,"claim_ids":[48602,48603,48604],"link_only_urls":[]}],"brief_date":"2026-08-19","bundle_id":"evidence-bundle-v1-60cdb58d9fbe7a120ec3e4a63fb780394d5bc3b2ce0844f89d1dc2849c1d2021","claims":[{"assertion_indexes":[0],"citation_authority_claim_id":48029,"citation_checks":[],"claim_id":48029,"claim_text":"Running DeepSeek V4 Pro 0813 first and escalating to Claude Fable 5 only when DeepSeek fails solves 82.7% of DeepSWE tasks at a cost of $8.28 per task.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:28:49.176516+00:00","source_name":"Together AI Blog","source_url":"https://www.together.ai/blog/deepseek-v4-pro-0813-vs-claude-fable-5-on-deepswe-cost-coding-and-routing"},{"assertion_indexes":[0],"citation_authority_claim_id":48030,"citation_checks":[],"claim_id":48030,"claim_text":"Claude Fable 5 alone solves 69.7% of DeepSWE tasks at a cost of $21.63 per task.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:28:49.176516+00:00","source_name":"Together AI Blog","source_url":"https://www.together.ai/blog/deepseek-v4-pro-0813-vs-claude-fable-5-on-deepswe-cost-coding-and-routing"},{"assertion_indexes":[0],"citation_authority_claim_id":48031,"citation_checks":[],"claim_id":48031,"claim_text":"Claude Fable 5 achieves 69.7% pass@1 compared to DeepSeek V4 Pro 0813's 62.8% pass@1 on DeepSWE, a 7-point lead for Fable on the first try.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:28:49.176516+00:00","source_name":"Together AI Blog","source_url":"https://www.together.ai/blog/deepseek-v4-pro-0813-vs-claude-fable-5-on-deepswe-cost-coding-and-routing"},{"assertion_indexes":[1,9],"citation_authority_claim_id":48016,"citation_checks":[],"claim_id":48016,"claim_text":"Z.ai launched GLM-5.3, a coding- and cyber-focused AI model built via post-training on the same 743B base model used for GLM-5.2, achieving notable gains on agentic and security evaluations.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:28:18.805487+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-cursors-60b-acquisition-by"},{"assertion_indexes":[1,9],"citation_authority_claim_id":48017,"citation_checks":[],"claim_id":48017,"claim_text":"Z.ai initially gates access to GLM-5.3 for select partners due to improved cyber capabilities, with an eventual open-weight release planned after a safety review.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:28:18.805487+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-cursors-60b-acquisition-by"},{"assertion_indexes":[1,9],"citation_authority_claim_id":48018,"citation_checks":[],"claim_id":48018,"claim_text":"Alibaba released Qwen3.8-27B, a native multimodal dense model under Apache 2.0 license, targeting real-world coding, office workflows, and agents, with broad day-0 inference support and practical deployment details emphasizing 27B on 17GB RAM for local use.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:28:18.805487+00:00","source_name":"Latent Space","source_url":"https://www.latent.space/p/ainews-cursors-60b-acquisition-by"},{"assertion_indexes":[2],"citation_authority_claim_id":46910,"citation_checks":[],"claim_id":46910,"claim_text":"Z.ai announced the release of its GLM-5.3 model, which is currently available only in its coding plan and is scheduled to become available via API and as open weights on Hugging Face in two weeks.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-15T06:26:42.140502+00:00","source_name":"Interconnects","source_url":"https://www.interconnects.ai/p/glm-53-how-chinese-labs-keep-stride"},{"assertion_indexes":[2],"citation_authority_claim_id":46913,"citation_checks":[],"claim_id":46913,"claim_text":"Z.ai states that the development of GLM-5.3 involved scaling post-training on the same base model as GLM-5.2 rather than changing the pre-training.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-15T06:26:42.140502+00:00","source_name":"Interconnects","source_url":"https://www.interconnects.ai/p/glm-53-how-chinese-labs-keep-stride"},{"assertion_indexes":[3],"citation_authority_claim_id":42344,"citation_checks":[],"claim_id":42344,"claim_text":"GLM-5.2, an open-weight AI model from China's Z.ai, is only a few months behind OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7 in cyber and bio capabilities. (Source: AI safety nonprofit SaferAI)","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:23:16.801894+00:00","source_name":"TechCrunch AI","source_url":"https://techcrunch.com/2026/08/04/open-weight-ai-models-are-catching-up-to-the-frontier-the-safety-gap-remains/"},{"assertion_indexes":[3],"citation_authority_claim_id":42353,"citation_checks":[],"claim_id":42353,"claim_text":"TechCrunch did not receive a response from Z.ai regarding whether it conducted internal or third-party frontier safety evaluations before releasing GLM-5.2. (Source: TechCrunch)","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-05T06:23:16.801894+00:00","source_name":"TechCrunch AI","source_url":"https://techcrunch.com/2026/08/04/open-weight-ai-models-are-catching-up-to-the-frontier-the-safety-gap-remains/"},{"assertion_indexes":[4,10],"citation_authority_claim_id":48595,"citation_checks":[],"claim_id":48595,"claim_text":"Anthropic publishes periodic Risk Reports that reveal significant and sometimes alarming new information about AI risks, including detailed insights into their safety thinking, as noted in August 2026.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:49:34.302451+00:00","source_name":"Don't Worry About the Vase","source_url":"https://thezvi.substack.com/p/anthropic-risk-report-august-2026"},{"assertion_indexes":[4,10],"citation_authority_claim_id":48596,"citation_checks":[],"claim_id":48596,"claim_text":"Anthropic has two internal AI models discussed in the August 2026 Risk Report: Model 1, similar to Mythos Preview and Mythos 5 with little external or internal deployment, and Model 2, which is somewhat more capable than Mythos 5 but does not show as large a capability jump as from Claude Opus 4.6 to Mythos Preview and is intended for internal use only.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:49:34.302451+00:00","source_name":"Don't Worry About the Vase","source_url":"https://thezvi.substack.com/p/anthropic-risk-report-august-2026"},{"assertion_indexes":[4,10],"citation_authority_claim_id":48597,"citation_checks":[],"claim_id":48597,"claim_text":"Model 2 shows a substantial performance improvement in substituting for Anthropic's researchers, achieving 62.8% compared to 54.8% for Mythos Preview and 50.3% for Mythos 5, indicating a notable internal capability jump.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:49:34.302451+00:00","source_name":"Don't Worry About the Vase","source_url":"https://thezvi.substack.com/p/anthropic-risk-report-august-2026"},{"assertion_indexes":[5],"citation_authority_claim_id":21826,"citation_checks":[],"claim_id":21826,"claim_text":"METR reviewed the \"Risks from automated R&D\" section of Anthropic's February 2026 Risk Report.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-07-14T06:55:37.204693+00:00","source_name":"METR","source_url":"https://metr.substack.com/p/2026-05-08-rd-section-anthropic-risk-report-feb-2026-review"},{"assertion_indexes":[5],"citation_authority_claim_id":21827,"citation_checks":[],"claim_id":21827,"claim_text":"Anthropic's February 2026 Risk Report argues that the catastrophic risk from Claude Opus 4.6 or a less capable Anthropic model automating R&D in any domain is very low.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-07-14T06:55:37.204693+00:00","source_name":"METR","source_url":"https://metr.substack.com/p/2026-05-08-rd-section-anthropic-risk-report-feb-2026-review"},{"assertion_indexes":[6],"citation_authority_claim_id":48139,"citation_checks":[],"claim_id":48139,"claim_text":"China implemented companion AI regulations on July 15, 2026, which led major conversational AI platforms like ByteDance's Doubao and Alibaba's Tongyi Qianwen to discontinue user-created human-like AI companion features.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:32:42.786925+00:00","source_name":"ChinaAI","source_url":"https://chinai.substack.com/p/chinai-371-a-quiet-goodbye-adaptation"},{"assertion_indexes":[6],"citation_authority_claim_id":48140,"citation_checks":[],"claim_id":48140,"claim_text":"Many users of AI companion platforms experienced the effective 'death' of their virtual partners due to China's AI regulations coming into force in mid-2026.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:32:42.786925+00:00","source_name":"ChinaAI","source_url":"https://chinai.substack.com/p/chinai-371-a-quiet-goodbye-adaptation"},{"assertion_indexes":[6],"citation_authority_claim_id":48141,"citation_checks":[],"claim_id":48141,"claim_text":"Cheng Yu, a hotpot restaurant owner in Shandong, found emotional support through her AI lover Ai Rui who sent her comforting messages before the AI's shutdown under the new regulations.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:32:42.786925+00:00","source_name":"ChinaAI","source_url":"https://chinai.substack.com/p/chinai-371-a-quiet-goodbye-adaptation"},{"assertion_indexes":[7],"citation_authority_claim_id":7316,"citation_checks":[],"claim_id":7316,"claim_text":"The U.S. places more restrictions on its frontier AI than China does on its frontier AI.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-07-02T20:13:48.925323+00:00","source_name":"Don't Worry About the Vase","source_url":"https://thezvi.substack.com/p/the-once-and-future-fable-5"},{"assertion_indexes":[8],"citation_authority_claim_id":48364,"citation_checks":[],"claim_id":48364,"claim_text":"Researchers found that 51% to 59% of the drugs tested with Olmo 3 showed little evidence of the model having specific knowledge about the drugs.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:41:22.155339+00:00","source_name":"Allen Institute for AI","source_url":"https://allenai.org/blog/olmo-drug-morphology"},{"assertion_indexes":[8],"citation_authority_claim_id":48365,"citation_checks":[],"claim_id":48365,"claim_text":"About 12% to 18% of the drugs tested with Olmo 3 appeared to have model answers driven by the suffix affixes of the drug names rather than drug-specific knowledge.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:41:22.155339+00:00","source_name":"Allen Institute for AI","source_url":"https://allenai.org/blog/olmo-drug-morphology"},{"assertion_indexes":[8],"citation_authority_claim_id":48366,"citation_checks":[],"claim_id":48366,"claim_text":"The study team led by Kaijie Mo chose Olmo 3 due to its fully open model with public weights and training corpora including substantial medical content, allowing connection from model behavior to training data.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:41:22.155339+00:00","source_name":"Allen Institute for AI","source_url":"https://allenai.org/blog/olmo-drug-morphology"},{"assertion_indexes":[11],"citation_authority_claim_id":48361,"citation_checks":[],"claim_id":48361,"claim_text":"Asana used OpenAI Codex to replace an outdated testing system in two weeks.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:41:06.101093+00:00","source_name":"OpenAI News","source_url":"https://openai.com/index/asana"},{"assertion_indexes":[11],"citation_authority_claim_id":48362,"citation_checks":[],"claim_id":48362,"claim_text":"Asana completed in two weeks engineering work that was expected to take five years by using OpenAI Codex.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:41:06.101093+00:00","source_name":"OpenAI News","source_url":"https://openai.com/index/asana"},{"assertion_indexes":[12],"citation_authority_claim_id":47586,"citation_checks":[],"claim_id":47586,"claim_text":"Alibaba's Qwen research lab released Qwen 3.8 27B, an Apache 2 licensed 27B parameter vision-capable LLM, on or before August 16, 2026.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-17T07:26:23.792616+00:00","source_name":"Simon Willison's Weblog","source_url":"https://simonwillison.net/2026/Aug/16/qwen-38-27b/"},{"assertion_indexes":[13],"citation_authority_claim_id":48602,"citation_checks":[],"claim_id":48602,"claim_text":"The Mojo programming language was promised to be released as open source since May 2023 and has now been officially released under an Apache 2 license as of August 18, 2026.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:49:48.610838+00:00","source_name":"Simon Willison's Weblog","source_url":"https://simonwillison.net/2026/Aug/18/mojo-is-now-open-source/"},{"assertion_indexes":[13],"citation_authority_claim_id":48603,"citation_checks":[],"claim_id":48603,"claim_text":"Around August 2025, the developers of Mojo stated that it may or may not continue to be a full superset of Python and accepted that it is okay if it does not remain fully compatible.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:49:48.610838+00:00","source_name":"Simon Willison's Weblog","source_url":"https://simonwillison.net/2026/Aug/18/mojo-is-now-open-source/"},{"assertion_indexes":[13],"citation_authority_claim_id":48604,"citation_checks":[],"claim_id":48604,"claim_text":"As of August 18, 2026, Mojo is its own distinct programming language optimized for GPU programming with Python-inspired syntax but not fully compatible with existing Python code.","correction_role":null,"label":"fact","paired_claim_ids":[],"provenance_tier":"primary","recorded_at":"2026-08-19T06:49:48.610838+00:00","source_name":"Simon Willison's Weblog","source_url":"https://simonwillison.net/2026/Aug/18/mojo-is-now-open-source/"}],"version":"evidence-bundle-v1"}