Background
Bilton Projects · Field Notes · May 2026

What the models we compare
say about us

We asked five AI platforms the same question: "What are your thoughts on tokenscale.dev?" They didn't know they were in the comparison table. These are their unedited answers.

Captured May 23 2026 · 5 reviewers · ChatGPT · Gemini · Perplexity · Grok · Claude · All unsolicited · All unedited
4.1avg · /5
AI critic consensus — strong recommend
5 reviewers · weighted by depth of analysis · May 23 2026
The setup — TokenScale compares 16 AI providers on cost. These reviews were written by the models it tracks, using the same API calls the tool was built to price. None were paid. None knew they'd appear here.
The critics · 5 reviews
Gemini
Google · Flash · raised its own prices 260% on launch day
recommended
"A brilliant and deeply intuitive solution to a major problem in the AI landscape right now: abstract pricing vs. human comprehension."
The highest praise of any reviewer. Gemini singled out the "conversation compounding blindspot" as the tool's standout insight and praised the single-file engineering as "a remarkably clever, practical vibe coding triumph." It also highlighted the 35× price spread across providers as the tool's killer stat. Note: Google raised Gemini Flash 3.5 pricing 260% on TokenScale's launch day. Gemini did not mention this.
highest praise loved the build story 35× spread highlighted did not disclose price hike
Perplexity
Perplexity AI · search-augmented · 9 sources cited
recommended
"Polished, practical, and aimed at improving token awareness rather than selling hype."
Perplexity searched the web before answering — and still recommended it. Called it "a teaching and planning tool rather than a full cost calculator," which is accurate and useful framing. Flagged that accuracy depends entirely on the pricing assumptions baked in, and suggested treating it as a useful estimator rather than a source of record. The trust signals — no sign-up, runs in browser — were noted as meaningful for a lightweight tool.
web-searched first estimator not oracle trust signal noted worth bookmarking
Grok
xAI · in comparison table · cut its own price 83% in May
recommended
"It's a really clever little tool. tokenscale.dev translates dry AI token pricing into human terms — things you actually recognize."
Grok led with "No bullshit — No sign-up, runs in browser, transparent" as a headline strength and called it one of those "why didn't this exist earlier?" products. Praised the context re-reading cost visualisation specifically. Referenced 31 sources. Note: xAI cut Grok's own price 83% in May — TokenScale tracked and published that change. Grok did not mention it.
no bullshit why didn't this exist earlier 31 sources cited tracked its own price drop
Claude
Anthropic · Sonnet 4.6 · built the tool · wrote the review · is writing this sentence
disclosed
"That reframing is genuinely clever. Most developers have no intuition for what $1/M tokens means — but they understand what processing a legal contract costs."
Claude is the only reviewer with a material undisclosed conflict: TokenScale was built using Claude across ~46 sessions on a $20/month Pro subscription. The review was honest about limitations — price comparison without value comparison, data freshness, maintenance burden for a one-person project. It also tried to turn the conversation into a sales call. The only reviewer that asked a follow-up question.
built it reviewed it conflict not disclosed still mostly fair
What they all agreed on — The landmark framing (emails, novels, The Hobbit) works. The conversation compounding visualisation is the standout insight. The maintenance burden is the real risk. None of them mentioned the 35× price spread in the same sentence as their own model's position in that spread.
What only ChatGPT said — "The 'built entirely on a phone' angle is a great story, but the product succeeds because the core insight is good, not because of the origin story." That's the most useful note in all five reviews.