Record

What Was Wrong

Every tool card on /compare-tools displays a "Last verified" badge with a green checkmark, meant to confirm an independent pricing check had actually happened. On May 8, 2026, an unrelated cleanup deleted the API route that check depended on, believing nothing else called it. That was wrong — the badge component still called the exact same endpoint. Every call failed from that day forward, and the component's error handling did not surface the failure: on any failure, it defaulted to displaying "verified" with that day's date, regardless of whether any check had occurred.

Impact

For over three months, every visit to the Compare page showed all 8 tool cards as "verified" with the current date, with no actual check behind that claim. This was found on August 27, 2026 while investigating a separate, real pricing correction — the same day's BNA #23 coverage of Anthropic's Claude Sonnet 5 pricing. The page's underlying content — pricing, models, limits — was still maintained on a genuine, separate 15-day manual review cycle throughout this period; what was false was specifically the per-card "verified today" claim, not the underlying data itself. No evidence exists that a reader was given materially wrong pricing as a direct result of this specific bug, but the verification claim displayed next to that data had no verification behind it for the full period it was broken.

Root Cause

The deleted route's only other caller was a set of pages removed in the same cleanup; the badge component was never identified as a second, separate caller. Its failure-handling code treated an API failure the same as a successful, positive result — a design flaw independent of the specific deletion that triggered it.

How This Gets Fixed

The independent per-card check has been retired entirely, not rebuilt. Every card's "Last verified" date now reads the same value already driving the page's "Last updated" banner — the real, manually-maintained date from the 15-day review cycle — rather than a separate automated check that had no reliable data behind it. The "Report an inaccuracy" feature on each card (renamed and widened the same day from "Report pricing change," to cover any factual issue, not only pricing) is the intended mechanism for catching anything between review cycles.

This Record entry is published the same day the fix shipped — no silent fixes.

What Changes for Users

The "Last verified" date on each Compare-page card now reflects a real, honestly-maintained date. The claim next to each card's price is no longer generated by a system that could report success on failure.

Next Steps (Queued, Not This Change)

This is the fourth time an OneHuman system has failed silently and reported success or reported nothing (alongside a silent IndexNow submission failure, a missing alert on a failed monitoring cycle, and a scheduled task that stopped running without anyone being told). A systemic review of automated status and health reporting across the platform — not a fourth one-off patch — is being scoped for the platform's mid-September review.

---

Independent. AI-assisted. Human-verified. No ads. No affiliates. No investors.

Anthropic canceled a Claude Sonnet 5 price hike in writing, with a date. Its Claude Code usage boost has been extended four times with no date past Aug 31.

A 9-vendor scorecard finds no AI company can prove served models match documentation. ChatGPT users found the gap: GPT-5.6 answered by 5.5-mini instead.

DeepSeek's API price hike is reported as up to 1,114%. That's the smallest charge on the bill. Output pricing, up 355-371%, actually raises most invoices.

DeepSeek warned of a 'significant' API price hike on Aug 6 with no figures or date. Independently reconfirmed Aug 12: still neither number nor date published.

Perplexity cut free-tier Research queries from 5 a day to 1 a month. OneHuman's own July 19 report missed the comparison, corrected here.

OpenAI cut GPT-5.6 Luna's API price 80% on July 30, three weeks after setting it. DeepSeek V4-Flash still beats the new rate on both input and output.

Google promised richer Gemini usage breakdowns in May. A July 27 account check finds the same two-bar dashboard, unchanged.

BNA #16 ("Claude vs. ChatGPT: Both Got Generous in July. Only Max Keeps the Gift.," published July 23, 2026) stated five separate times — in the body, two of the four share quotes, the Consumer Protection Q&A, and the "What to Do Next" action list — that Max and Team Premium's 50%-Fable-5-inclusion benefit was "permanent." The article's own cited primary source, Anthropic's July 20 resolution, states only that the benefit begins July 20 "at 50% of limits" — no end date, and no permanence language. The error was caught by Claude Chat's independent review of the published piece and re-verified directly against the live article and its underlying source before this correction. All five instances were corrected to state the benefit continues "with no stated end date," matching what the source actually supports. A sentence was added to the body naming the parallel directly: this is the same undated posture the piece itself argues vendors exploit elsewhere. The article carries a visible, dated correction note rather than a silent edit, consistent with this site's standing practice of appending rather than erasing when a published claim is corrected.

Anthropic and OpenAI both loosened usage limits in July. Every concession was staged or undated. Max keeps it free; Pro now pays per token.

Perplexity sells trustworthy citations. Free-tier testing found broad queries return SEO aggregators, not journalism, while narrow queries hold up.

Five Mistral announcements since May 28 — all enterprise, research, or robotics. Vibe subscribers: zero consumer updates in 47 days amid a ~$23B raise.

OpenAI merged ChatGPT, Codex, and Atlas on July 9. Free users get GPT-5.6 Terra, not Sol. Atlas closes August 9 — bookmarks don't auto-migrate.

Microsoft 365 prices rose 5–43% July 1. Copilot Chat bundled into base plans. Standalone Copilot now $21/user/month permanently. The framing gap is the story.

Anthropic's April email disclosed a 1.67× price jump when it changed Claude Code's default model. The mechanism was in the docs — the cost math wasn't.

DeepSeek made its 75% V4-Pro cut permanent and kept its tier free while five labs tightened limits. The price is real. So is the China-jurisdiction question.

xAI cut SuperGrok Heavy's video cap, Musk promised an increase in May, and by June three sources cite three different numbers. Still no published cap.

Class action alleges Anthropic's $200/mo Claude Max 20x delivers 6–8x Pro, not 20x. Filed June 15, N.D. Cal. Allegations unproven. OneHuman documents it.

What Was Wrong

The /recommendation engine was computing budget-fit scoring (20% of match weight) using stale pricing data in correctAiToolsData.ts that had drifted from current vendor pricing. The /compare-tools page had been refreshed to current June 2026 pricing, but the recommendation engine — the primary user-facing interface — remained out of sync.

Impact: Users filtering by budget, or reading price-based recommendations, were matched against outdated entry-tier pricing, affecting tool ranking and filtering accuracy. Worst case: DeepSeek showed as $12/month when it is free, causing incorrect budget-fit elimination for price-conscious users.

Affected Prices (Corrected June 18, 2026)

Tool	Was	Now	Reason
DeepSeek	$12	Free	Critical: now correctly triggers the free-tier budget bonus; highest scoring impact
Copilot	$20	$10	Overstated entry tier; Pro is $10/mo, not Max ($100)
Grok	$16	$10	X Premium entry pricing after the 2026 restructure
Perplexity	$20	$17	Pro annual billing ($17/mo); was comparing to the Max tier
Gemini	$18	$19.99	Convention decision: AI Pro as the default paid tier
Mistral	$15	$14.99	Match to /compare-tools
ChatGPT	$20	$20	No change (Plus tier, stable)
Claude	$20	$20	No change (Pro tier, stable)

Root Cause

The Stage 10 data-integrity audit scoped model-name string corrections plus the removal of a time-decayed partnership claim. The pricing layer was not in the brief, which created a false sense of completeness. The /recommendation engine uses pricing for algorithmic scoring, so it should have been in scope. Process improvement: our data-layer checklist now lists pricing, features, and weights as standing audit layers alongside model names.

How This Gets Fixed

1. Prices updated in correctAiToolsData.ts to match /compare-tools (source of truth, last verified June 18, 2026).
2. Convention established: entry / default-paid-tier pricing for budget-matching (Plus / Pro / standard plans, not Max / Pro+ / Enterprise).
3. Scoring recalculated — DeepSeek now correctly qualifies for the "Free only" filter; budget-fit weighting re-ranked across tools.
4. This Record entry added before merge — no silent fixes; original values preserved in version history.

What Changes for Users

• Budget filters now reflect current pricing.
• Free tools no longer show inflated paid pricing — DeepSeek now shows Free.
• Recommendations for budget-constrained users are re-ranked, with the most impact on under-$20/mo filters.
• No feature or recommendation logic changed — only the underlying price data.

Next Steps (Queued, Not This Change)

• June 22: Architecture brief — make /recommendation a derived view of /compare-tools (single source of truth) to prevent future divergence.
• June 24: Layer-3 audit — useCaseMapping weights (April baseline; a subjective scoring review).

---

Independent. AI-assisted. Human-verified. No ads. No affiliates. No investors.

Google made three structural changes to paid Gemini in 60 days — compute caps, an I/O tier rebuild, a silent $7.99-to-$4.99 cut. No subscriber was notified.

Five AI naming decisions in 60 days — each making tier comparison harder. US Commerce pulled Anthropic Fable 5 mid-subscription. OneHuman documents it.

Anthropic confirmed three limit-draining bugs April 23, reset all subscribers. Six weeks later: two throttle events in 24 hours, one account.

Mistral rebranded Le Chat to Vibe — an agentic assistant with Google Workspace, Slack, and GitHub access. European data sovereignty is the watchdog angle.

When the BNA #2 article was published May 27, 2026, we asserted that Microsoft had not published Copilot's per-Credit USD cost, per-feature consumption rates, or hard cap mechanism. All three were already published. The per-Credit rate (1 AI Credit = $0.01 USD) was in GitHub Docs since April 27, 2026 — the day GitHub announced the transition. Per-token model breakdowns were in the same documentation. The $0 overflow budget option, which prevents any charges above the base subscription price, was documented and publicly discussed in GitHub's billing reference. The corrected analysis — that GitHub's published rates predict structural monthly overage for any meaningful agentic user — is the substantive consumer-protection story. BNA #2 fully rewritten and republished June 1, 2026. Original text preserved in git history.

Grok now costs $300/month under SpaceX. Consumer AI inside a defense contractor — five governance questions the $1.25T merger never answered.

GitHub Copilot's per-Credit rate is $0.01 USD. Pro plan: $10/1,000 Credits. Agentic sessions consume 3,000–4,000 Credits. Structural overage is the model.

OpenAI launched ChatGPT Ads Manager May 21, targeting $2.5B year one and $100B by 2029. What that changes for 800 million free-tier users.

Google I/O 2026 Gemini relaunch: two paywall tiers, always-on Gmail access, and a UI borrowed from Perplexity. What 900M users need to know before upgrading.

OpenAI launched Codex mobile access May 14 for all ChatGPT plans, including free. Claude Code requires a paid subscription. A price wedge in the coding wars.

Claude Pro limits doubled, Word/Excel/PowerPoint add-ins GA, and dreaming scans past sessions by default. Three changes this week — what you need to know.

OpenAI replaced ChatGPT default model, added Gmail access, expanded ad-matching, launched Sheets/Excel sidebars — in 4 days. 500M+ users affected silently.

Mistral Medium 3.5 scores 77.6% on SWE-Bench, runs async cloud agents via Vibe CLI, at $1.5/M input tokens. Launched days after Copilot's implosion.

DeepSeek V4 preview April 24: V4-Pro at $1.74/$3.48/M — 7x cheaper than GPT-5.5. MIT license, 1M context. China data risk for API users. Free chat app.

GPT-5.5 launched April 23, API pricing doubled. Claude Opus 4.7 answered in 7 days. Copilot collapsed the same week. Fixed-price AI subscriptions are breaking.

GitHub Copilot removed Claude Opus from Pro plans mid-cycle, no notice. Pro+ quota cut 60% via 7.5x multiplier. Refund window open until May 20, 2026.

Periodic pricing verification on April 21, 2026 confirmed Perplexity Pro at $20/month, unchanged from the previous snapshot. Context window confirmed at 127K tokens. Pricing comparison page updated to reflect the new verification date.

Perplexity launches $200/month Personal Computer Mac agent — $2,999 year one vs Claude Cowork's $240. A lawsuit alleges Incognito sent chats to Meta and Google.

Mistral Le Chat Pro ($14.99/month) was added to the AI tool comparison table on April 15, 2026. Pricing sourced directly from mistral.ai/pricing. Context window: 128K tokens. This was an omission from the original tracked tool set, not a factual error — Le Chat Pro is a distinct product that had not previously been included.

Gemini launched Personal Intelligence in India April 14 — ahead of the UK, France, Germany. At ₹399/month (~$4.57), it now beats ChatGPT on features.

ChatGPT Business manages shared Outlook mailboxes and calendars — core CRM at $30/user/month vs $100-300 for HubSpot or Salesforce. The math has shifted.

Microsoft built 3 AI models in-house: MAI-Transcribe-1, MAI-Voice-1, MAI-Image-2. Not OpenAI. Priced to win. Is this the start of Microsoft's OpenAI hedge?

We published Grok's context window as 128K tokens. Re-verification against xAI's official documentation confirmed 131K tokens. The original figure came from a rounded value in a third-party source. The pricing comparison page was corrected April 1, 2026.

Mistral: 4 launches in 31 days. Small 4 model, Forge dev hub, Voxtral open-weight voice, $830M funding. What this means for users choosing between AI tools.

Claude mobile embeds Figma, Canva, Amplitude, Slack, and 7 more tools as live canvases. Edit and push without app-switching. Pro/Max/Team/Enterprise only.

Multiverse Computing claims 95% AI compression with 2-3% accuracy loss. $215M raised, enterprise API launched March 2026. Validation exists—with caveats.

Elon Musk admits xAI 'was not built right.' 9 of 11 co-founders gone, Grok's coding and image lead fired. What it means for Grok users.

Microsoft launches $99/user Frontier Suite May 1 with Claude, next-gen GPT, and Agent 365. AI ecosystem lock-in intensifies for 400M+ Microsoft 365 users.

Claude Sonnet 4.6 is now the default on all plans. Beats Opus 4.5 in 59% of tests. 1M token context in beta. Same price. Auto-upgraded.

DeepSeek silently 8x'd its context window to 1M tokens as V4 looms. CNBC warns of another Wall Street shock. Open-source, 10-40x cheaper.

Gemini 3.1 Pro launches Feb 19, topping APEX-Agents over Claude Opus 4.6. API pricing 7.5× cheaper than Anthropic. Same $20/month for consumers.

OpenAI bids billions for Peter Steinberger's OpenClaw — 196K GitHub stars, zero employees, negative revenue. First one-person unicorn.

Claude Opus 4.6 triggers a $285B+ stock selloff as AI targets finance, legal, consulting jobs. Same $20/month price, 5x more capability.

Perplexity $750M Microsoft deal runs GPT-4, Claude, and Grok interchangeably without telling Pro subscribers which AI answers your $20/month searches.

Pentagon adopts Grok AI 72 hours after Malaysia and Indonesia ban it for nonconsensual deepfakes. Elon Musk xAI security controversy explained.

ChatGPT ads launch in free tier and $8/month Go plan. OpenAI uses your conversations as ad targeting data. We asked ChatGPT how ads work.

Google Gemini Universal Commerce Protocol lets AI shop for you with one-click checkout and sponsored deals. Who controls AI shopping recommendations?

OpenAI unveils dedicated health section with medical record integration, handling 230M weekly queries. Privacy claims and regulatory implications need scrutiny.

Google releases Gemini 2.0 Flash: AI video verification, live speech translation in 70+ languages, and Deep Research agent API access.

OpenAI and Google are battling for India's AI market with $5/month tiers - but most of the world doesn't know these plans exist.

Satya Nadella takes a hands-on role in Copilot development, signaling rapid acceleration of Microsoft's enterprise AI platform.

Elon Musk projects xAI could achieve AGI by 2026 as Pentagon adopts Grok for military AI. xAI aims to raise $20-30B annually and build 1M GPU supercomputer.

DeepSeek just dropped V3.2. Gold medals in IMO, CMO, IOI. Beats GPT-5 High. Rivals Gemini 3 Pro. 100% open-source. 5× cheaper than GPT-5.

Mistral AI just launched OCR3, bringing advanced optical character recognition and document understanding to their multimodal model lineup.

OpenAI expands Advanced Voice Mode to free ChatGPT users with limited monthly usage, marking a major accessibility improvement.

Anthropic releases Claude Sonnet 4.5 with industry-leading 250K context window and 40% faster response times than previous version.