The week in one sentence: The AI model market is moving faster than any calendar you can book time against — and three specific deadlines this week mean UK operators who don't make conscious decisions will have decisions made for them by default. Here is what the AIFA team is watching, and the three choices to make before Friday.
Signal 1: GPT-5.6 "Kindle" is coming this week
OpenAI's next model — codenamed "kindle-alpha" in Codex backend logs since May — is expected to launch by 28 June 2026 according to prediction market Polymarket, which gives it 83% odds. API pricing is expected at roughly one-third of current GPT-5.5 rates. That matters for UK SMBs because it makes GPT-class reasoning and coding quality more budget-accessible. A model that costs £1.60 per million input tokens instead of £5.00 changes the economics of automation workflows meaningfully — not overnight, but in planning terms.
The practical operator question is not "which model is best" — it is: does a cheaper, better model change your current workflow or budget decisions? If you are on GPT-5.5 for API-heavy tasks, a one-third price drop from a successor model is worth evaluating quickly. If you are using AI for one-off drafting, the difference is minimal.
The AI model price war between OpenAI, Anthropic, and Google is compressing in the direction that benefits UK businesses. Each new model generation tends to offer better performance at lower cost than its predecessor. GPT-5.6 appears to continue that pattern.
Signal 2: Copilot Wave 3 — agents inside M365
If your business runs on Microsoft 365, this is your lowest-friction route to agentic automation. Wave 3 is not a new product to buy or a separate tool to learn — it is an update to the apps your team already uses, on licences you are already paying for. The shift is from Copilot-as-assistant (suggest the next sentence) to Copilot-as-agentic-execution (complete a multi-step task: research, draft, format, and send — in a governed workflow without leaving Outlook). The inclusion of Claude Opus 4.7 alongside GPT 5.5 Instant means the M365 environment now has access to two leading models, not just Microsoft's own.
The strategic point here is that Microsoft has done the integration work. For UK businesses who have been watching AI from the sidelines because the tooling felt disconnected from their actual software, Wave 3 removes much of that friction. The entry point is already inside the apps your team uses every day.
Signal 3: Claude Fable 5 pricing is now live
Anthropic's Claude Fable 5 — the most capable publicly available model they have shipped — was free on all paid plans from 9 June. That window is now closed. Using Fable 5 from today requires usage credits. The evaluation window was 14 days. If you tested it, you have a basis for a decision. If you did not, the question now is whether to run a comparison between Fable 5 (credits) and Sonnet 4.6 (plan-included) on your highest-stakes workflow to decide whether the quality improvement justifies the cost. This is a business decision, not a technical one — and it is yours to make deliberately rather than by omission.
There is also ongoing disruption context: a US export control directive on 12 June affected Fable 5 API access, which was restored on 18 June with new identity verification for certain jurisdictions. UK users were not directly restricted, but the disruption underlines why evaluating models before pricing deadlines — not after — is worth building into how you manage AI tools.
Three decisions to make before Friday
Decision 1: Write your GPT-5.6 test case now
Do not wait for the model to arrive and then scramble to evaluate it. Identify the specific task you want to test (a proposal draft, a code function, a complex analysis), define what "better than current" looks like, and have it ready to run within 48 hours of the launch announcement. If you do not prepare the test in advance, you will defer evaluation — and that means not making the pricing and workflow decision you need to make.
Decision 2: If you are on M365, book time this week for Copilot Wave 3
Log in to Microsoft 365 Copilot Chat inside Word or Outlook. Look for the agentic task palette — multi-step automation tasks that go beyond single-prompt responses. Identify one workflow where agentic execution could save your team meaningful time: a proposal that requires research and formatting, a follow-up sequence that requires drafting and scheduling, a data summary that requires pulling from multiple sources. You do not need to deploy anything. You need to see what is now possible inside the tools you already pay for.
Decision 3: Run a Fable 5 vs Sonnet 4.6 comparison on your highest-stakes workflow
If you missed the free window, you can still evaluate Fable 5 by purchasing a small credits allocation. Pick the workflow where AI output quality has the highest business consequence — a client-facing document, a complex analysis, a response to a difficult query. Run it on both models. If Fable 5 delivers a material improvement, calculate your monthly token usage and work out what the credit cost would be. If the improvement is marginal, Sonnet 4.6 is your daily driver for zero additional cost. This comparison takes less than an hour and gives you a grounded decision rather than a guess.
The pattern across all three signals is the same: the AI market is moving at a pace that rewards businesses who make deliberate, timely decisions — and penalises those who defer. None of these decisions require large commitments. They require 60–90 minutes of focused attention this week.
Read the full breakdowns for each story below — GPT-5.6 Kindle, Copilot Wave 3, Claude Fable 5 pricing, and the London Mayor's £12m AI fund for UK SMBs.