Apodex open-sourced a Smol SFT series (0.8B, 2B, 4B) alongside its 35B-A3B mini. Their reported evaluation shows the 4B SFT outperforming open 30B-class models on both BrowseComp and BrowseComp-ZH. This kind of result keeps pointing to the same pattern in deep research: model size is not always the main constraint. What remains to be tested is whether the same gap persists on longer-horizon research tasks, where retrieval, planning, and evidence aggregation matter more.

Three Credible Sources, Three Different Timelines: I Asked Apodex @Apodex_AI to Judge the Foldable iPhone Rumors   As of July 16, 2026, Apple has not confirmed whether it will release a foldable iPhone before the end of the year.   The problem is not a lack of reporting. It is that credible reports point in different directions.   Bloomberg's Mark Gurman says Apple's first foldable iPhone remains on track for a September 2026 debut. Nikkei and Reuters describe engineering problems that could delay shipments. DigiTimes and MacRumors suggest production has slipped, but Apple is still targeting fall 2026.   That made the foldable iPhone a useful test for Apodex.   I asked whether Apple would both announce and begin selling its first foldable iPhone by December 31, 2026. Apodex had to identify conflicting sources, explain their weights, build three scenarios, state a confidence level, and list the signals that would invalidate its conclusion.   The distinction between "announce" and "begin selling" matters. Apple could introduce the phone in September while delaying availability until December or 2027.   How Apodex Weighed the Sources   Bloomberg's September timeline became the baseline. Gurman has a strong record on Apple product timing, and other outlets independently support a fall 2026 target. Apodex still down-weighted the claim because the timing was not final and later reporting introduced production risk.   Nikkei's engineering details were treated as credible and given substantial weight. However, Apodex did not treat a 2027 delay as the base case because the report described it as a worst-case outcome, not a confirmed schedule change.   The Barclays view received less weight because it came from a single analyst note. Still, its pattern was plausible: Apple introduced the iPhone X in September 2017 and released it in November. A September announcement followed by December sales could reconcile the reports.   What Actually Decides the Outcome   The forecast depends on hinge reliability, OLED and assembly yield, production speed, and Apple's quality threshold.   A product can be ready to announce while remaining difficult to manufacture at scale. Better yield supports fall sales; continued instability makes December or 2027 more plausible.   Three Scenarios   Apodex divided the outcome into three paths:   - Early case, about 25%: Apple announces the phone in September and begins sales in late September or October. - Base case, about 55%: Apple announces it in September, with limited retail availability beginning in December. - Delayed case, about 20%: engineering or yield problems push consumer sales into 2027.   The base case preserves Bloomberg's September introduction while accommodating the reported production delay and December-shipment forecast.   What Would Prove It Wrong?   The probability of a 2026 sale should fall sharply if Bloomberg, Reuters, Nikkei, or Apple reports that volume production has moved into 2027. The same applies if suppliers delay components into Q1 2027 or if assembly and display-yield failures continue into October.   The cleanest public test will be Apple's September event. If it passes without a foldable-iPhone announcement, the base case fails. Confirmed mass production, carrier preparation, or 2026 delivery dates would move the forecast in the opposite direction.   Why the Test Matters   The useful result was not the percentage itself. It was the structure of the judgment.   Apodex compared contradictory claims, assigned different weights, built multiple paths, and stated what would force it to change its mind. That matches its official positioning as a heavy-duty solver: turning information into evidence through verification and reaching a defensible conclusion under uncertainty.   The narrower claim is more useful: when the answer does not yet exist, Apodex can make the evidence, uncertainty, and failure conditions inspectable.   Try Apodex:

AI reliability can't come from "self-reflection" alone. Welcome to the era where a separate agent audits the answer before you get it 🔬 Title: Apodex-1.0: A Verification-Centric Agent Team for Discoverative Intelligence URL: 🔬 Overview A system that shifts from a single-agent reasoning loop to a verification-centric distributed agent team. In heavy-duty mode it becomes an asynchronous team that specializes, cross-checks, and audits its own evidence before answering. ❓ Challenges Solved Reliability on hard, open-ended problems can't come from a model's parametric memory alone. The premise: the hardest research problems are bounded not by model capacity but by what the model is allowed to interact with. 💡 Methodology & Proposed Approach ・A main agent asynchronously spawns specialized sub-agents with independent contexts and tools ・A shared report pool aggregates parallel findings without blocking on slower tasks ・A verification agent team handles conflict resolution, fact-checking, and draft review ・The core idea is verification as external audit: the reasoning agent and auditing agent are separated, and the verifier is free to disagree ・It coordinates up to 150 sub-agents over 15,000+ steps in a single task 📊 Experimental Results ・BrowseComp 90.3 / DeepSearchQA 94.4 / BrowseComp-ZH 84.1 ・FrontierScience-Research 46.7 (+8 vs competitors) / SuperChem 74.2 (+12 over next-best) ・Heavy-duty mode lifts the base by +14.8 on BrowseComp and +18.4 on FrontierScience-Research ・The open-source 4B-SFT beats every 30B-class open-source model on BrowseComp #AIAgents# #DeepResearch#