OpenAI takes back the enterprise growth lead
For much of 2026, the enterprise AI narrative favoured Anthropic. The company reportedly reached a $65 billion revenue run-rate and posted $11.6 billion in Q2 revenue, surpassing OpenAI in some business segments and building serious momentum among coding teams. Then GPT-5.6 Sol shipped on July 9, and the picture flipped.
According to spending data from corporate card platform Ramp, business API spending on OpenAI grew 82% quarter-over-quarter in Q3, edging out Anthropic's 76%. OpenAI itself reports total revenue growth of 35% for the quarter, with enterprise revenue up more than 50% — a sharp re-acceleration after Anthropic's brief lead, as The Decoder reports.
One caveat before we canonise the growth numbers: percentage growth is not the same as absolute revenue. Anthropic's 76% may still represent more dollars depending on the base. But the direction of travel is unmistakable — and it is exactly the trend line Anthropic does not want to show investors right now.
GPT-5.6 Sol: a model, and a pricing strategy
GPT-5.6 Sol is the flagship variant of the GPT-5.6 family (Sol, Terra and Luna tiers), launched on July 9, 2026. The commercial push, however, is not just about raw capability. OpenAI cut GPT-5.6 Sol API pricing by more than 20% for a three-month window, bringing input tokens to $4 per 1M and output tokens to $20 per 1M — a 33% reduction on the output side.
For European developers paying in euros, that works out to roughly €3.70 per 1M input tokens and €18.50 per 1M output tokens at current exchange rates. This is our own conversion, so treat it as indicative — but the point stands: frontier-model output pricing at this level was unthinkable for a GPT-class model a year ago.
Benchmarks: where Sol actually stands
On the Artificial Analysis Coding Agent Index, GPT-5.6 Sol scores 80, compared with 77 for Anthropic's Claude Fable 5 — at an estimated cost of about $1.04 per completed task. That combination of score and price is exactly the metric we care about in our own AI Arena benchmarking: not just "who is smarter," but "what does the smarter answer actually cost?"
Here is how the current frontier pricing compares (promotional prices where applicable):
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Coding Agent Index |
|---|---|---|---|
| GPT-5.6 Sol (promo) | $4.00 | $20.00 | 80 |
| Claude Fable 5 | n/a | n/a | 77 |
| Gemini 3.7 Flash | $0.75 | $3.75 | — |
| Grok 4.6 | $2.00 | $6.00 | — |
| DeepSeek V4-Flash | $0.14 | $0.28 | — |
Claude Fable 5's standard API pricing was not disclosed in the source data, so the table focuses on the score where it matters. For pure price, DeepSeek remains in a league of its own. For the top of the coding-agent leaderboard, Sol's temporary pricing makes it the value pick right now.
The European angle: availability, AI Act, and the euro effect
GPT-5.6 Sol is available in the EU through OpenAI's standard API offering — there are no regional blocks for European companies. The regulatory picture, however, has changed meaningfully over the past year. GPT-5.6 is a general-purpose AI (GPAI) model, which means it now falls under mandatory transparency disclosures under Article 50 of the EU AI Act and the binding GPAI Code of Practice. The voluntary transition frameworks of 2025 are gone, replaced by active EU AI Office enforcement.
In practice for a European company: if you deploy Sol in a customer-facing product, you need the usual transparency documentation, AI-generated content disclosures, and an appropriate governance trail. For most B2B SaaS operators that is paperwork layered on existing GDPR obligations, not a blocker. The larger unresolved issue remains data residency — sending sensitive data to US-hosted APIs still demands a DPIA, and for many European enterprises, contractual commitments on data handling.
The price war itself is genuinely good news for European teams. API costs are denominated in USD, so a 20%+ cut applies whether the bill is paid from Prague, Berlin or Paris. And the competitive pressure keeps alternatives honest: Mistral's open-weight Leanstral 1.5 and Mistral Medium 3.5 remain viable European options, while DeepSeek V4-Flash anchors the low end of the market.
Why now? The IPO factor
Anthropic is reportedly preparing for an IPO, and the timing of OpenAI's temporary price reduction is almost certainly not coincidental. A three-month promotional window on a flagship model directly pressures the revenue projections Anthropic will present during its roadshow. It is a commercial shot across the bow, aimed precisely where investors look first: growth rates.
What should European buyers watch? When the three-month promo ends around October, OpenAI has two options: keep the price and accept margin pressure, or revert and risk a migration wave. The smart play is already clear — build multi-model routing that can switch between Sol, Claude, and open-weight models without re-architecting your pipeline. That way, whoever wins the price war, the buyer wins too.
Is GPT-5.6 Sol available to European companies?
Yes. OpenAI offers GPT-5.6 Sol through its API across the EU. Because it is a general-purpose AI model, EU deployments fall under the AI Act's Article 50 transparency obligations and the binding GPAI Code of Practice — in practical terms, documentation and disclosure duties for providers and deployers.
How long does the temporary price cut last?
The more-than-20% reduction is a three-month promotion: $4 per 1M input tokens and $20 per 1M output tokens (a 33% cut on output). It runs until approximately October 2026, after which prices may revert to pre-promo levels.
Should we switch from Claude to GPT-5.6 Sol?
On the Artificial Analysis Coding Agent Index, Sol scores 80 versus 77 for Claude Fable 5, at roughly $1.04 per task. But benchmark scores are only one input — evaluate against your own workloads, and consider multi-model routing so you can follow pricing and quality wherever they move next.
