Anthropic will release bigger, more expensive models than those released so far.

The claim is that Anthropic will release models bigger (and more expensive) than any it had released by 28 Feb 2025. TRUE: Anthropic releases a model that it describes as larger, or that credible third-party analysis or pricing shows to be larger and more expensive, than its largest model up to that date (Claude 3 Opus and Claude 3.7 Sonnet). Examples would be a model priced above Claude 3 Opus's tier, or one stated to use more parameters or compute. MOSTLY TRUE: the model was released and is clearly larger, but its price is equal to the previous top tier rather than higher. MOSTLY FALSE: a larger model was only announced, or was made available only to a restricted internal or partner group. Also mostly false if only more capable but not bigger or more expensive models were released. FALSE: dated evidence shows Anthropic stopped scaling, or its subsequent releases are all at or below its earlier size and price tiers.

Set before the research began.

1. E1 Updated Anthropic model comparison table Simon Willison's Weblog · 22 May 2025

   “Opus 4 matches the pricing of the older Opus 3 - $15/million for input and $75/million for output.”

2. E2 Introducing Claude 4 Anthropic · 22 May 2025 Own record

   “Pricing remains consistent with previous Opus and Sonnet models: Opus 4 at $15/$75 per million tokens (input/output) and Sonnet 4 at $3/$15.”

3. E3 Anthropic in numbers: 71 sourced changes AI Adoption · 20 October 2026

   “FLOP: its largest training run with a published estimate, Claude 3.7 Sonnet (Feb 2025)”

4. E4 Claude Fable Anthropic · 1 September 2026 Own record

   “Claude Fable 5.1 is priced at $10 per million input tokens and $50 per million output tokens.”

5. E5 Anthropic in numbers: 71 sourced changes AI Adoption · 20 October 2026

   “FLOP: its largest training run with a published estimate, Claude 3.7 Sonnet (Feb 2025)”

6. E6 Anthropic in numbers: 71 sourced changes AI Adoption · 20 October 2026

   “20 Jun 2024OutputNewClaude 3.5 Sonnet released, trained with an estimated 2.7×10²⁵ FLOP”

7. E7 Anthropic in numbers: 71 sourced changes AI Adoption · 20 October 2026

   “Training compute, cost and hardware are Epoch's estimates, shown only where it has one.”

8. E8 Anthropic in numbers: 71 sourced changes AI Adoption · 20 October 2026

   “4 Mar 2024Claude 3 Opus frontierAPI or hosted access———”

9. E9 Over 30 AI models have been trained at the scale of GPT-4 Epoch AI · 30 January 2025

   “Claude Opus 4 | 1.5e26 | Speculative | Sonnet 3.5 was likely trained on around 1-4e25 FLOP, similar to GPT-4. Sonnet 4 may be scaled up from Sonnet 3.5. Opus 4 was probably trained on between 5e25 to 2e26 FLOP.”

10. E10 Over 30 AI models have been trained at the scale of GPT-4 Epoch AI · 30 January 2025

    “Claude 3 Opus | 1.6e25 | Low-precision | Compute imputed from benchmark scores. Known to be similar in training compute to GPT-4.”

11. E11 omsi/sources/anthropic.com/claude-fable-5-1-system-card/2026-10-26.md at master · wsxiaoys/omsi GitHub (wsxiaoys/omsi wiki; mirrors Anthropic system card) · 1 September 2026 Own record

    “It is also more cost-efficient than its predecessors on many evaluations, matching or exceeding Fable 5 at roughly half the cost per task on agentic coding benchmarks.”

12. E12 Over 30 AI models have been trained at the scale of GPT-4 Epoch AI · 30 January 2025

    “was not trained in any way that involved a larger or more expensive model”

Claude Opus 4 was released at the same price as the earlier Opus tier [E1][E2]. Epoch AI's speculative estimate puts its training compute above Claude 3 Opus [E9][E10]. No evidence item gives a compute figure for Claude 3.7 Sonnet, and a 2026 tracker still lists Claude 3.7 Sonnet as Anthropic's largest training run with a published estimate [E3], so the evidence does not show that a later model is larger than both benchmark models.