Reporting tells you what happened. This page tells you what we are waiting to find out. Every item below is an open question raised by a story we have already published, with the source that raised it.
Nothing here is a prediction. These are dated, checkable questions with a specific answer coming, and we will update this page when each one resolves.
Pricing that is scheduled to change
Does Claude Sonnet 5 rise by 50 percent on 31 August 2026?
Anthropic lists USD 2.00 input and USD 10.00 output per million as an introductory rate, ending 31 August 2026, after which it becomes USD 3.00 and USD 15.00. Anyone modelling costs on the current figure is modelling the wrong one. See the Claude Sonnet 5 profile.
Does DeepSeek raise its API pricing?
DeepSeek’s own pricing page warns that an overall increase is expected, described as significant. At USD 0.14 per million input tokens it is currently the cheapest capable model in the directory, and that is the basis on which a lot of people are choosing it. See the DeepSeek-V4-Flash profile.
Commitments that have not landed yet
Does Alibaba release the Qwen3.8-Max open weights?
Alibaba announced Qwen3.8-Max on 3 August 2026 with a commitment to publish open weights within a week. At the time of our last check we could not confirm they had shipped, so we do not describe the model as open. If they arrive, the strongest reported agentic coding model becomes something you can run yourself. See the Qwen3.8-Max profile and our piece on where open weights moved after Llama.
Does Gemini 3.1 Pro lose its preview label?
Google’s reasoning flagship launched on 19 February 2026 and still ships under a preview model identifier. Google has confirmed Gemini 3.5 Pro is in testing and Gemini 4 is only in pre training. This is the clearest single test of whether the August leadership split changes Google’s shipping cadence. See our analysis and the Gemini 3.1 Pro profile.
Figures we could not verify
Do OpenAI’s two pricing pages agree?
OpenAI’s GPT-5.6 launch post and its developer pricing page quote different rates for Terra and Luna. We use the docs pricing page because that is the one OpenAI maintains for billing, and we say so on both profiles rather than picking one silently. See the Terra and Luna profiles.
Does Anthropic publish the Claude Max 20x price?
The Max tier is quoted only as from USD 100 per month across a 5x and a 20x option. The price of the 20x tier is not published, so we record it as unverified. See the Claude profile.
Do Canva and CapCut publish a global price list?
Neither vendor made current consumer pricing readable to us. CapCut’s own help article states that cost varies by region, device, and promotion. We would rather print unverified than a number we could not read on the vendor’s own page. See the Canva and CapCut profiles.
Claims waiting on independent evidence
Does anyone reproduce Qwen3.8-Max’s benchmark scores?
Alibaba reports 86.6 on Terminal-Bench 2.1, the strongest agentic coding figure any lab has published. Every number comes from its own release materials. That is normal for a new model and it is not a criticism, but a self reported score from any lab is a claim until someone else reproduces it. See the Qwen3.8-Max profile.
How this page works
An item is added when a published 18BYTE story raises a question with a specific, checkable answer arriving later. It is removed when the answer arrives and the relevant profile or story is updated. We do not add items we cannot source, and we do not use this page to trail stories we have not reported.
If something here has resolved and we have not caught it, tell us at hello@18byte.com. See also our editorial standards and corrections policy.