Which AI assistant is worth your $20 a month?

The one to pick if you do serious written or technical work, and the reason is what comes bundled rather than what happens in the chat window. Claude Pro includes Claude Code alongside chat, which means the same twenty dollars covers a coding assistant you would otherwise buy separately — for anyone who writes code even occasionally that is the whole argument. On output, it holds a voice across long documents better than anything here and stays consistent deep into an argument rather than drifting into summary. Two things to know before committing. API pricing moves on 1 September 2026: Sonnet 5 goes from $2 to $3 per million input, and a tokenizer change lands at the same time adding 10 to 35 percent more tokens on code, so the effective rise on coding workloads is larger than the sticker. And the $100 and $200 Max tiers exist purely for people hitting limits, so do not buy one on principle.

The default, and defaults are worth something. ChatGPT has the widest integration surface, the most third-party tooling built against it, and the deepest bench of features that exist because enough people asked. If your work spans many small varied tasks rather than one deep one, breadth beats peak capability and this wins on breadth. The tier structure is also the most sensible here: Go at $8 is a genuine option for light use rather than a crippled trial, and it is the plan most people paying $20 out of habit should probably be on. Image generation via GPT Image 2 is included, which quietly removes another subscription. What you give up is depth at the top end on long-form writing, where Claude is better, and on real-time information, where Grok is.

The cheapest way to get a capable assistant, and the most capable free tier of the six. Google cut AI Plus to $4.99 in June 2026, which makes it the only serious sub-$5 option in the category and the right answer for anyone whose usage is a few times a week. Context length is the technical argument: Gemini handles very long documents without the chunking gymnastics other tools require, and if your work involves pasting large reports or codebases that difference shows up immediately. The bundle is the other argument — storage, Gmail and Workspace integration come attached, so the effective value is higher than the line item if you already live there. Against that, it trails Claude on sustained written quality and ChatGPT on third-party tooling, and Ultra at up to $249.99 is the steepest ceiling here.

The only one with live access to X, and that is a genuine capability rather than a marketing line. If your work involves breaking news, market sentiment, or what people are saying right now, no amount of general capability substitutes for a current feed. Grok 4.6 also scores 61 on the Artificial Analysis Intelligence Index, matching GPT-5.6 Sol, so you are not trading capability for recency. The pricing is where care is needed. SuperGrok at $30 is 50 percent more than the $20 standard, the free tier is capped around 10 messages per two hours, and Heavy at $300 is the steepest consumer plan in the category. On API there is a specific trap: cross 200,000 input tokens and the entire request reprices from $2 and $6 to $4 and $12 per million — not the overage, the whole request.

The value pick, and the one most Western buyers have not looked at. Kimi meters in credits rather than messages, and the detail that matters is that K2.6 in chat is free for all users and does not draw on your credit pool at all — so the free tier is genuinely usable rather than a trial. Moderato at $19 is the first tier with K3 access and undercuts every $20 competitor. Where it separates is the agent tooling: Allegretto at $39 adds Agent Swarm and Kimi Code at 5x, which is functionality the $20 Western tiers do not offer at any price. On API, K3 at $3 and $15 per million undercuts Claude Opus 5 at $5 and $25 and GPT-5.6 Sol at $5 and $30 on both sides. The caveats are ecosystem size and a credit model that takes a read-through to understand — and note the 1M context window sits on the higher tiers, not on Moderato.

Here for one reason and it is a strong one: at roughly $0.27 and $0.28 per million tokens, DeepSeek V4 Flash is an order of magnitude cheaper than anything else on this page. If you are running high-volume classification, extraction or summarisation where per-token cost dominates and peak reasoning does not, that difference is the entire decision. Chat is free to use. Two things to hold in mind. Prices moved sharply this year — the August 2026 rate is up 93 percent from $0.14, so treat the current figure as a snapshot rather than a floor. And deepseek-chat and deepseek-reasoner are deprecated from 24 October 2026, so anything you build now needs a migration path. As a daily driver for chat it is the weakest here; as a cost line in a pipeline it is unbeatable.