DeepSeek spent the better part of 2026 being the reason "AI coding is basically free" felt true. Then, at 16:00 UTC on August 16, its API prices jumped — up to 371% higher on output tokens, with peak/off-peak billing replacing the old flat rate. DeepSeek didn't get expensive in absolute terms, but it stopped being the free ride that made every other coding plan look overpriced by comparison.
That changes the question worth asking — and it's worth asking narrowly. Skip the resellers, the multi-model credit bundles, and the IDE wrappers that route through someone else's API key. Compare only the subscriptions sold directly by the labs that actually build the models: OpenAI, Anthropic, and Google. On that narrower comparison, using Artificial Analysis Intelligence Index data against real subscription prices, one plan comes out ahead on cost per task: ChatGPT Plus.
TL;DR: what to pick
| Question | Answer |
|---|---|
| Cheapest per coding task? | ChatGPT Plus ($20/mo) — GPT-5.6 Sol's Medium reasoning tier posts the lowest cost per task of the three |
| Highest raw intelligence score? | Claude Pro ($20/mo) running Claude Opus 5 at Max reasoning effort |
| Best value if you want both? | ChatGPT Plus and Claude Pro cost the same $20 — pick by whether you optimize for cost per task or peak intelligence |
| Where does Google AI Pro land? | Cheapest list price ($19.99/mo) but a worse cost-per-task and intelligence trade than ChatGPT Plus for coding specifically |
| Is DeepSeek's API still cheaper than these? | For high, steady token volume — yes, per-token. For casual/bursty use, a flat subscription now wins on simplicity |
| Why no Cursor, GLM, or multi-model plans here? | They resell or route access to these same underlying models at their own markup — this comparison covers only the labs' own direct pricing |
Why this comparison only counts the three frontier labs
A lot of "cheapest AI coding plan" comparisons mix in resellers — IDE subscriptions, multi-model credit bundles, open-weight hosting plans — that route requests through someone else's API key at their own margin. That's useful information, but it answers a different question than "what does OpenAI, Anthropic, or Google actually charge for their own model."
This comparison sticks to subscriptions sold directly by the lab that trained the model:
| Lab | Plan | List price | Model on the plan |
|---|---|---|---|
| OpenAI | ChatGPT Plus | $20/mo | GPT-5.6 Sol, all reasoning-effort tiers |
| Anthropic | Claude Pro | $20/mo | Claude Opus 5 & Sonnet 5, all reasoning-effort tiers |
| Google AI Pro | $19.99/mo | Gemini 3.7 Flash, High reasoning tier |
Same $20-ish price across all three. The difference is what that money buys per task.
ChatGPT Plus: cheapest per task
GPT-5.6 Sol on ChatGPT Plus scales cleanly across Medium → High → Xhigh → Max reasoning effort, and its Medium tier posts the lowest cost per coding task of any plan in this three-way comparison on the Artificial Analysis Intelligence Index — cheaper per task than Claude Pro's comparable tier and cheaper than Google AI Pro's Gemini 3.7 Flash tier, despite all three plans costing roughly the same per month. If your priority is minimizing cost per completed coding task rather than chasing the single highest intelligence score, ChatGPT Plus is the direct answer.
Claude Pro: highest intelligence, higher cost per task
Claude Opus 5 on Claude Pro's higher reasoning-effort tiers (High, Xhigh, Max) posts the top Artificial Analysis Intelligence Index scores of the three plans — Max reasoning effort is the single highest score in this comparison. That intelligence comes at a real cost: each step up in reasoning effort raises the cost per task, and Claude Pro's top tier costs meaningfully more per task than ChatGPT Plus's cheapest usable tier. Same $20/month unlocks every setting, so the trade-off is entirely about which reasoning effort you actually need for the work in front of you.
Google AI Pro: cheapest sticker price, not the best coding value
Google AI Pro lists at $19.99/month — technically the lowest of the three — but on pure coding cost-per-task, Gemini 3.7 Flash's High reasoning tier runs higher than GPT-5.6 Sol's cheapest tier while posting a lower intelligence score. That's not necessarily a knock on Gemini generally — it's built to be fast and cheap for a broad range of tasks, not to top a coding-intelligence leaderboard — but for coding specifically, it's the weakest of the three plans on both axes. See explainx.ai's full Gemini 3.7 Flash vs. Grok 4.6 vs. Sonnet 5 vs. GPT-5.6 comparison for where it fits outside pure coding.
Why DeepSeek's price hike actually matters here
DeepSeek isn't a flat-fee subscription, so it isn't in the table above — it's an API you pay per token. Its role in this conversation was as the baseline: as long as DeepSeek's tokens stayed dirt cheap, any $20/month plan had to justify itself against "just use the API and pay pennies."
That baseline moved. DeepSeek confirmed peak/off-peak pricing increases up to 371% on output tokens starting August 16, 2026 — some third-party trackers reported increases as steep as 1,100% on specific token types during peak hours. DeepSeek's new rates are still below GPT-5.6 Sol and Claude Fable 5 per-token pricing, so it hasn't stopped being cheap in absolute terms. But the gap that made "just use the API" an obvious default has narrowed, which is exactly why a direct frontier-lab subscription comparison is worth running now instead of assuming the API route always wins.
What this means if you're learning to code with AI
If you're new to using AI coding assistants day to day, the reasoning-effort setting matters more than the brand name. Both ChatGPT Plus and Claude Pro let you dial the same model up or down, and dialing up costs real money per task without a guaranteed proportional intelligence gain. Learning when Medium effort is enough (most edits, small refactors) versus when Max effort earns its cost (multi-file architecture decisions, gnarly bugs) is a skill in itself, and one explainx.ai covers as part of loop engineering — running AI coding agents efficiently rather than maxing out settings by default.
If you're choosing your first paid plan and coding is the primary use case, ChatGPT Plus is the lowest-cost-per-task entry point among the frontier labs' own subscriptions; Claude Pro is the upgrade path once you hit tasks where extra reasoning effort clearly pays for itself.
Related reading
- DeepSeek V4 Prices Just Went Up — Does It Really Match GPT-5.6?
- What Your $20 AI Subscription Actually Costs the Company
- Which AI Subscription Is Actually Worth Paying For?
- Gemini 3.7 Flash vs. Grok 4.6 vs. Sonnet 5 vs. GPT-5.6 Sol Comparison
- Fable 5 vs. Grok 4.6 vs. GPT-5.6 Sol vs. Qwen3.8-Max: Who Actually Wins?
- Loop Engineering for Coding Agents: The Complete Guide
- Artificial Analysis Intelligence Index — explainx.ai Dictionary
Prices and Intelligence Index scores reflect what was published as of August 17, 2026. Subscription pricing changes often — check each lab's own pricing page before committing to a plan.