Three announcements in July were read as one piece of news: AI was getting cheaper. On the 7th, Accenture Edge and Google Cloud launched a line of pre-configured agents for mid-sized companies. On the 21st, Microsoft signed a multi-year, multi-billion dollar commitment funding Mistral's compute infrastructure in Europe. On the 24th, Anthropic shipped Claude Opus 5 at the previous generation's price. Seven weeks and a full release cycle later, September's price sheets say something else.
AI adoption in the enterprise is the organisational work that turns technical access to artificial intelligence tools into real professional use: scoped use cases, integrated company data, access governance and change management measured over time, work whose cost answers to its own calendar and staffing constraints.
Model prices no longer move in one direction
The 2024 and 2025 pattern fitted in a sentence: at equal capability, the price fell year over year. The Token Price Index published by BenchLM, which tracks 40 models across three segments, no longer measures anything of the kind. In its 3 September 2026 edition, the frontier segment is up 72% year over year and the budget segment up 36.8%. Its third segment, mid-tier, rests on a rate that does not exist.
Anthropic's catalogue carries no price increase at all, and it shows where the index gets it wrong. Claude Opus 5, released on 24 July 2026, holds the Opus 4.8 price: 5 dollars per million input tokens, 25 on output. Claude Fable 5.1, released on 1 September 2026, keeps Claude Fable 5's 10 and 50 dollars unchanged, and cuts cache reads to a quarter of the previous rate. On 10 August 2026, Anthropic cancelled the Claude Sonnet 5 increase it had announced for 1 September and made the introductory rate the standard one, 2 dollars on input and 10 on output (Anthropic public price list, 10 September 2026). Yet the section of BenchLM's 3 September 2026 edition that lists what moved credits two of those moves to Anthropic: Claude Fable 5.1 entering the frontier segment at 20.00 dollars per million blended, and a Claude Sonnet 5 mid-tier increase from 4.00 to 6.00 dollars per million. That second move is the very increase cancelled three weeks earlier. The index therefore still carries a price that does not exist, with no published erratum: its mid-tier segment should be set aside until it is recomputed, while its other two segments remain usable.
At the entry level, the rebound has a traceable cause: a share of those rates were launch promotions, and they expire. Z.ai's promotional pricing for GLM-5.3-Flash ended on 9 September 2026 and the list rate doubled (Z.ai public price list, 10 September 2026). Upstage's discount on Solar Pro 4 does not stop at once, it steps down: 90% until 11 September 2026, then 70% until 10 October, and full price after that, at 0.30 dollar per million input tokens and 1.20 on output (Upstage public price list, 10 September 2026). Moving from one step to the next, the same volume therefore bills three times higher.
DeepSeek illustrates the other form of instability, with no promotion involved. On 16 August 2026 the provider moved to peak and off-peak pricing that doubles its V4 rates from 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. On 10 September 2026 a new V4.1 Flash price sheet took effect, at 0.15 dollar per million input tokens and 0.60 on output, off-peak. And from 14 September 2026, calls to V4-Pro are routed to V4.1 Flash and billed at the Flash rate (DeepSeek public documentation, 10 September 2026). Three billing changes in a month across the same catalogue.
For a finance function, the consequence is direct. The token price has become an unstable input: it climbs at the top, it climbs back at the entry level the day a promotion lapses, and a single catalogue's billing rules can change three times in a month.
Price cuts have not gone away either, and they arrive with the same notice, which is to say none. On 30 July 2026, OpenAI cut the price of Luna, the cheapest model in its GPT-5.6 range, by 80% (OpenAI announcement, 30 July 2026). The cancelled Sonnet 5 increase is the same mechanism running the other way. What changed since 2025 is predictability: a rate can now shift either way, on a specific model identifier, between two budget cycles.
July's two other announcements complete the picture. Microsoft's 21 July 2026 press release announced the distribution of Mistral's models through Foundry and Copilot Studio: since then, Mistral Medium 3.5 and Mistral OCR 4 have become available in Foundry, and Medium 3.5 in Copilot Studio, with deployment that covers cloud, cloud-connected and fully disconnected environments for regulated industries. Accenture Edge and Google Cloud target, in their 7 July 2026 announcement, companies with revenues between 300 million and 3 billion dollars. Compute is now negotiated in both directions, and basic technical configuration is becoming standardised.
Why don't model prices decide AI adoption?
Because the price of a model and a company's cost of adoption rest on unrelated bases. The first is negotiated in dollars per million tokens and now moves both ways. The second is measured in redefined roles and teams trained for professional use, and it has held the same level for two years.
S&P Global finds that 46% of generative AI pilots are abandoned before reaching production. MIT puts at 95% the share of pilots that never reach measurable return on investment. The 2026 edition of Deloitte's State of AI in the Enterprise report puts at 21% the share of companies deploying agents with governance rated as mature. Gartner expects roughly 40% of agentic AI projects to be cancelled by 2027. All four measures describe how work is organised.
- Redefining roles: deciding what an agent handles alone, what a human validates, what stays off-limits.
- Integrating company data: connecting the systems actually used in production.
- Training teams for professional use: moving from an isolated personal prompt to a scoped and repeatable use case.
- Installing governance: monitoring agents continuously once they are running.
These four workstreams take the same amount of leadership, IT and business-team time, whatever the list price of the underlying model.
The regulatory calendar has already moved
On 20 July 2026, France's data protection authority, the CNIL, published a joint exploratory note with the French AI and Digital Council on agentic AI, systems that act on a user's behalf. On 2 August 2026, the transparency obligations under Article 50 of the EU AI Act became applicable. They fall under Article 99(4), which caps fines at 15 million euros or 3% of global annual turnover. The 35 million euro or 7% ceiling, widely quoted in the press, sits in Article 99(3) and covers the prohibited practices of Article 5. The two regimes are distinct.
A company deploying cheaper agents without governance in place accumulates the same risk with more agents in circulation. The cost of governing was untouched by any of the summer's price moves.
What should a company do now?
When a technology gets cheaper, the reflex is budgetary: cut the envelope and treat the AI question as settled once the tool is installed. September's market makes that reflex expensive. An AI agent in production sized on promotional pricing becomes an unpredictable line item the day the promotion lapses, and it remains, without proper scoping, a system nobody is really supervising.
The budget freed up by a cheaper model is worth more elsewhere: on mapping use cases and on governing agent access. Those items have the advantage of being predictable. Their cost can be planned over a financial year, which no model price sheet still allows.
Before adding one more agent to the stack, a company benefits from knowing where it actually stands on these four workstreams. Koneetiv's AI maturity assessment measures technical foundations and governance in a few minutes, the two dimensions that decide whether an agent deployment holds up when the model price changes direction.