SciSpace Review

My first query went into the main search box: 

“What are the main approaches to sequence transduction without recurrence?” 

I wanted a real literature question, not a one-line definition, to see how the platform would handle it.

The search runs across several databases at once

Rather than pulling from a single index, SciSpace searched its own corpus, its full-text layer, Google Scholar, and arXiv in one pass. It also checked my personal library, which was empty, so nothing returned from there. The run gathered 84 papers into a results table.

The scoring is what caught my attention. Every paper carried a relevance score out of 100, a relevance tag, and a short reasoning note on why it matched my question. The top result sat at 100/100 with a one-line justification about attention-based encoder-decoder models. For anyone who checks their sources, that reasoning column removes a step.

The report arrives with a table and real citations

SciSpace then wrote a structured report titled non_recurrent_sequence_transduction. It opened with a TL;DR, moved into a section on architectural families for non-recurrent transduction, and set out a comparison table pairing each architecture with its main mechanism and its key advantage. The body text carried inline numbered citations, and the report closed with 8 references complete with DOIs.

I did not take those citations on trust. I clicked reference [1], and it opened the actual paper, “Self-Attentional Models Application in Task-Oriented Dialogue Generation Systems,” hosted on acl-bg.org with the original URL sitting in the address bar. The citation was genuine, not a fabricated entry.

Exporting the report is where the free plan stops

Reading the report costs nothing. Keeping it does. The instant I clicked to download, whether as Markdown or as PDF, SciSpace blocked me with an “Upgrade to download agent files” box and pushed its paid plans. On the free tier I can generate a full report and read it on screen, but I cannot carry a copy away.

Chatting with a PDF

Next I uploaded a 240-page PDF, the AP Biology Course and Exam Description, and used the chat panel on the right. I typed “Conclusions from the paper” and SciSpace returned a sectioned summary with each claim tagged by inline citations such as [1] and [2] that point back into the document. A highlight option lets me select any statement inside the PDF and drop it straight into the chat, and an “Explain math & table” button handles dense pages.

One feature here is locked. The Podcast option, which turns a paper into audio, sits behind the upgrade wall and does not work on the free plan.

Image generation works, saving it does not

SciSpace also generates images. I gave it the prompt “create an image of a futuristic cinematic landscape with glowing neon lights, stunning details, vibrant colors, and a realistic 3D aesthetic,” and it produced a detailed neon cityscape, then offered a second variation.

The finished image displayed in full inside the app. The same pattern held as with the report: I could look at it, not download it on the free plan.

A large library of ready-made agents

The Agent Gallery lists 2,867 agents grouped by task and by field, with task categories such as Literature Review and Deep Search, plus domain filters covering physics, chemistry, biomedical, and more. Featured agents include a Systematic Literature Review workflow and a PubMed search agent. Browsing them is free, while running them and exporting their output falls under the same credit and paywall rules as everything else.

What the free plan actually costs

My usage page tells the honest story. After three tasks I still held 92 of my 100 credits, resetting on 30 November. The history listed the cost per task: 3 credits for the dropout-and-BLEU query and 2 for the image. The sequence-transduction report also took 3.

So the common claim that the free credits vanish within a day did not hold up for me. At 2 to 3 credits per task, 100 monthly credits cover a solid amount of searching and reading. The free plan’s real constraint sits elsewhere. It lets me work on screen and read full outputs, yet it locks every download and a handful of features like Podcast behind a paid upgrade.