If you signed up for ChatGPT, Claude or Gemini a month ago, or picked a coding assistant off a benchmark chart from back then, the deal you locked in may already be out of date. AI token prices have fallen more than 21% in six weeks, according to The Decoder, even as usage keeps climbing. New moves from Microsoft on coding and Google on its own Gemini line make the same point from different angles: price and performance are moving faster than most subscription cycles.
The price floor keeps dropping
The Decoder's reporting lands on a simple number: token prices, the per-use cost behind every AI subscription and API call, dropped more than 21% in six weeks. That is happening industry-wide, not at one vendor. The practical effect is that a plan priced as competitive a month or two ago can now carry a markup over what the same capability costs elsewhere today.
The Decoder also points to a concrete gap worth checking before you pick a plan: OpenAI's models are priced noticeably cheaper per token than Anthropic's. For buyers comparing ChatGPT against Claude on cost alone, that is the comparison to start from.
Microsoft's coding model undercuts the leaders on price
Microsoft is marketing its new in-house coding model as a cut-price alternative to OpenAI's flagship, The Next Web reports: Microsoft claims 98% of that model's coding performance at well under half the price. Microsoft also claims its model nearly matches Anthropic's Claude Opus 5.5 on coding benchmarks while costing a small fraction as much.
Those are Microsoft's own benchmark figures, not independent test results, and coding benchmarks do not always predict how a model performs on a specific team's codebase. But the target is unambiguous: Claude Opus 5.5 is the model many coding-assistant buyers default to, and Microsoft is pricing explicitly against it.
Gemini's newest release outpaced its own predecessor within weeks
Google supplies the clearest evidence that today's best deal will not stay that way for long. The Decoder reports that Google's newest Gemini release outpaces its own predecessor, launched only weeks earlier, on both speed and benchmarks, while costing less. The predecessor was itself a flagship pick when it launched. Buyers who chose it for its price-to-performance ratio are now a generation behind on both counts, from the same company.
Price isn't deciding the market yet
None of this means the cheapest token price automatically wins. The Decoder notes that despite OpenAI's cheaper per-token pricing, Anthropic still commands the majority of token spending, a sign that price alone isn't driving buyer choice yet. Tom's Guide points to what else is: Claude has built a reputation for natural, consistent prose, a reason some buyers choose it over ChatGPT specifically for writing-heavy work. ChatGPT's counter is breadth: it bundles voice conversation features that matter to buyers who want one assistant for many tasks rather than a writing specialist.
That split, specialist quality against bundled breadth, sits on a separate axis from price. It is why the cheapest model per token is not automatically the right pick, and why brand loyalty is a weaker starting point than matching the model to the job.
What to check before you renew
The practical shift is from picking a brand once to matching a model to the task, and re-checking the comparison periodically given how fast the field is moving. Before renewing or signing up for ChatGPT, Claude, Gemini or a coding assistant, check:
- What you actually use it for. Writing-heavy work, broad day-to-day use and coding each favor a different model, per Tom's Guide and The Next Web's benchmark claims above.
- Whether your current price still reflects the market. Token prices are down more than 21% in six weeks, per The Decoder, so a plan that looked competitive recently may no longer be.
- Whether a benchmark claim is independent or vendor-supplied, as with Microsoft's coding-model figures, and whether it has actually been tested against your own workload.
No model holds a durable price or performance edge for long. The safest assumption is that today's best deal gets beaten within weeks, sometimes by the same vendor. Re-check price and benchmarks before renewing, and weigh specialism, writing, breadth or coding, over brand habit when the gap is close. Compare Robots' buying guides and help-me-choose tool cover how to weigh these tradeoffs case by case, and our AI news coverage tracks the releases that keep shifting the comparison.
Sources: The Decoder, The Next Web, Tom's Guide
Written with the help of AI tools, from published sources, under our editorial rules. Illustration generated with AI. No commission on purchases. Learn more


