The bottom line
- Cheapest way to the #2 model: Google AI Plus at $4.99/mo gets you Gemini 4 Argon — the same model as the $19.99 Pro plan and the $100+ Ultra plan.
- Best coder and best writer: Claude Opus 5.5, on Claude Pro for $20/mo. It also leads on research.
- Best researcher: Claude Opus 5.5 (67.9% on Humanity's Last Exam, Diamond).
- Images and video: no neutral test scores exist yet for any current model. Claude can't generate images at all, and ChatGPT has no video any more.
Each card shows the plan, the model it includes, and six everyday tasks. Bars run against the best subscribable model for that task — a full bar means best-in-guide, not perfect. A dash (—) means no verified public score exists for that model, not zero. Scores: coding = LMArena Coding Elo, writing = LMArena Creative Writing Elo, research = Humanity's Last Exam (Diamond) %, general = LMArena overall Elo — all Oct 28 – Nov 6, 2026 snapshots. Full sources at the bottom.
Free — $0/mo
Every provider has a free tier. None of the free models is separately tested by the neutral boards, so these cards show what you get, not a score.
Google Gemini
Free plan
Model: Flash-Lite
A smaller model than the paid plans — fine for quick questions, not the one the boards test.
OpenAI ChatGPT
Free plan
Model: GPT-6 (version not stated)
OpenAI doesn't name the model version on the free plan page, so no board score can be attached to it.
Anthropic Claude
Free plan
Model: Sonnet & Haiku (versions not stated)
The free tier runs smaller Claude models whose versions Anthropic doesn't state.
xAI Grok
Free plan
Model: Not stated
xAI doesn't name the model version on any of its plan pages.
Perplexity
Free plan
Model: "Basic AI models" (Perplexity picks)
Perplexity's free tier is a search tool first — it picks a basic model for you.
Meta AI
Free plan
Model: Muse Spark (Thinking mode: 1.1)
No neutral scores are published for Muse Spark 1.1 — the tested 1.3 powers Sean's assistant instead (see below).
Everyday — $5 to $30/mo
This is where the value is. The $4.99 Google plan runs the same model as the $19.99 one, and the best coder on the board costs $20.
Google AI Plus
Google · best value
Model: Gemini 4 Argon
Same model as the $19.99 Pro plan — you're paying for higher limits, not a better model.
Google Gemini AI Pro
Model: Gemini 4 Argon
Same model as Plus ($4.99). Pro adds higher limits and features — and it's the plan Sean uses for heavy research.
OpenAI ChatGPT Plus
OpenAI
Model: GPT-6 (version not stated)
OpenAI doesn't name the model version on the Plus plan page, so no board score can be attached to it.
Anthropic Claude Pro
Anthropic · best coder & writer
Model: Claude Opus 5.5
The best coder, the best writer, and the best researcher in this guide — for $20/mo. Fable 5.1 is available via usage credits only.
Perplexity Pro
Perplexity · you pick the model
Model: Your pick — GPT-6.1 Sol, Gemini 3.8 Flash, Claude Sonnet 5.5, Grok 4.7, Kimi K3, GLM 5.3, Nemotron 3.5
Scores shown are the best pick for each task — the model you choose changes the score.
xAI Grok Premium
xAI
Model: Grok 4.7 (flagship; plan page doesn't name the version)
The weakest flagship on nearly every board — and the plan page won't say which version you get.
Meta One Core
Meta
Model: Muse Spark (model not named on plan page)
More usage of the same Muse Spark — the plan page doesn't name the model version.
Power user — $100 to $500/mo
Mostly you're buying higher limits and the top creative model — not a better general chatbot. The $4.99 plan above runs the same Argon as Ultra.
Google Gemini AI Ultra
Model: Gemini 4 Argon
Same model as Plus ($4.99) and Pro ($19.99) — Ultra buys limits and extras, not a better model.
OpenAI ChatGPT Pro
OpenAI
Model: GPT-6 Astra
*Astra tops the other writing test (EQ-Bench Creative Writing, #1) while ranking #40 on Arena Creative — the two tests disagree; see notes. Its general score (1476) is well below the 1520 figure from a compiled table that circulated earlier.
Anthropic Claude Max / Ultra
Anthropic
Model: Claude Fable 5.1 (Opus 5.5 also included)
Fable is the creative-writing flagship — but Opus 5.5 (the better coder and writer) is also included on Max, so this tier's best model is the $20 Pro one.
Perplexity Max
Perplexity · you pick the model
Model: Your pick + Max-only — GPT-6 Astra, Gemini 4 Argon, Claude Opus 5.5, Claude Fable 5.1, Model Council
The only plan that lets you use the top model for every task — but the best picks are all available cheaper elsewhere (Opus 5.5 on Claude Pro $20, Argon on Google Plus $4.99).
xAI Grok Super / Ultra
xAI
Model: Grok 4.7 (flagship; plan page doesn't name the version)
Same flagship as the $30 Premium plan — the higher tiers buy limits, not a better model.
Sean's stack — not a subscription
The model behind this site's assistant, shown for comparison. It isn't something you subscribe to.
Muse Spark 1.3
Meta · powers the Muse assistant
Model: Muse Spark 1.3 (Max)
*The only model with a published neutral DeepSearchQA score (89.4%) — every other model is unmeasured on that test, so it isn't a ranking. Artificial Analysis Intelligence Index 48.1 (v4.3.2, Oct 15, 2026). No public MMLU-Pro, GPQA, or SWE-bench results exist for it — the dashes on the old version of this page were honest, not missing data.
- How to read this: scores come from neutral third-party tests, not the providers' own claims. Coding = LMArena Coding Elo, writing = LMArena Creative Writing Elo, general = LMArena overall Elo (Oct 28, 2026 snapshot, style control on, official lmarena-ai dataset). Research = Humanity's Last Exam, Diamond (Artificial Analysis, Nov 6, 2026). Bars run against the best subscribable model for each task. A dash means no verified public score exists, not zero.
- Correction: GPT-6 Astra's general score is 1476 (#37) on the verified Oct 28 LMArena board — not 1520, a compiled-table figure that circulated earlier. An earlier version of this page showed 1520.
- Not in any plan: Gemini 4 Carbon (1528, the top-rated model overall) isn't in any subscription — it's Fairwind-program only. The best model you can actually buy is Gemini 4 Argon (1520).
- The writing tests disagree: GPT-6 Astra is #1 on EQ-Bench Creative Writing v3 (2176.9) but #40 on Arena Creative Writing (1450). EQ-Bench v3 is judged by Claude Sonnet 4.6; the card shows the Arena figure.
- Research is the thinnest category: no neutral BrowseComp score exists for any model (only OpenAI's vendor-reported 91.5% for Astra), and only Muse Spark 1.3 has a neutral DeepSearchQA score. Humanity's Last Exam boards disagree with each other (Opus 5.5: 67.9% / 61.4% / 55.0% across the three) — this page uses the Artificial Analysis Diamond board throughout.
- Coding cross-checks: SWE-bench Verified figures (Opus 5.5 84.6%, Fable 83.8%, Astra 82.4%, Sol 82.3%, Argon 81.9%) come from one Oct 2026 compilation — the official board has accepted no new submissions since Feb 2026. Terminal-Bench 4.0 varies by harness (Vals vs the official Harbor board differ).
- Images & video: no neutral GenEval, DPG-Bench, or VBench scores exist for any current image or video model. Sora was discontinued (app closed Apr 2026, API shut down Sep 2026); Veo 3.1 is no longer in the Gemini app. Claude can't generate images or video at all.
- Plan pages that don't name the model: ChatGPT Plus, all xAI plans, and Meta One don't state which model version you get — those cards say so rather than guessing.
- Pricing: provider pricing pages, Nov 6, 2026. Gemini AI Ultra is listed as $99.99–$199.99 (tiered); ChatGPT Pro as $100 / $200 / $500.