Opus 5.5 Is Crazy Good and GPT-6-Sol Launched Too

Video thumbnail: Opus 5.5 Is Crazy Good and GPT-6-Sol Launched Too
Sep 22, 202618m 17s video lengthMatt Wolfe

The Signal

Anthropic and OpenAI launched new flagship models on September 22, 2026, marking a significant shift in AI economics. While both companies prioritized lower costs and faster speeds, Anthropic’s Claude Opus 5.5 is positioned as a tangible performance leap, whereas OpenAI’s GPT6 Soul and Luna models appear to be incremental updates focused on efficiency.

The Case

Anthropic Opus 5.5

  • Anthropic launched Claude Opus 5.5, a model priced at $4 per million input tokens and $20 per million output tokens, which the company claims runs 40% cheaper than its predecessor, Opus 5.0:32
  • Opus 5.5 is framed as the new state-of-the-art across key benchmarks, including a 66.4% score on agentic coding and a 57.8% score on Cursor Bench, though it remains contested whether these gains translate to broad real-world superiority.1:07
  • The most compelling evidence cited is not benchmark data but a series of public demonstrations, including playable game clones like Mario Maker and complex Blender animations, which show interactive capabilities significantly more advanced than models from a year ago.5:02

OpenAI GPT6 Series

  • OpenAI introduced GPT6 Soul and Luna on the same day as Anthropic’s launch, significantly reducing API costs—Soul dropped from $4 to $2 per million input tokens—to keep the company competitive in a fast-moving news cycle.9:08
  • Despite the price cuts, GPT6 Soul is portrayed as less capable than Anthropic’s Opus 5.5 and OpenAI’s own previous flagship, GPT6 Astra, with few public demos available to showcase a clear competitive advantage.
  • Economic efficiency varies by task; while Opus 5.5 is cheaper per token, its higher token consumption—roughly 119,000 tokens per task compared to 31,000 for GPT6 Soul—means that total per-task cost remains a critical metric for developers.13:44

The 1 Minute Signal Take

The choice between these models now rests on the trade-off between total cost-per-task and demonstrated creative capability. If you are a developer, prioritize testing your specific workloads against both Anthropic's efficiency-driven benchmark gains and OpenAI's lower token-usage model rather than relying on aggregate leaderboard rankings.

Pro Analysis

Why It Matters

The simultaneous release of these models marks a transition from a 'performance at any cost' race to a 'price-performance' race. Anthropic is effectively commoditizing high-end intelligence, forcing OpenAI to move from being the undisputed leader to a contender that must compete on value.

Strategic Implications

The divergence between token-cost and cost-per-task is the most critical strategic insight for any business integrating these models. Decision-makers must now model total cost of operation rather than relying on per-million-token marketing sheets.

Evidence & Hype Audit

The content relies heavily on benchmarks provided by the vendors and community-driven 'beauty benches.' While these are useful, they are prone to selection bias. The reliance on demos from early-access users provides a helpful, if anecdotal, look at capabilities, but lack the rigour of long-term production testing.

Counterarguments

Critics might argue that these benchmarks are 'cherry-picked' to favor specific strengths of the models. Furthermore, the 'beauty bench' as a measurement of model intelligence is notoriously subjective and may not correlate with actual business-critical reasoning capabilities.

Who Should Care

  • CTOs: To evaluate API migration strategies to lower operational costs.
  • Product Leads: To assess if the new creative capabilities allow for new product features.
  • Data Scientists: To investigate if the reported benchmark gains hold up under internal, domain-specific testing.

What to Do Next

  • Run an internal cost-per-task analysis for your most frequent LLM queries.
  • Conduct a side-by-side comparative test of Opus 5.5 and GPT6 Soul on your actual production codebases.
  • Wait for broader, non-early-access user reports before committing to a full-scale API migration.
  • Track the upcoming Friday news cycle to see if further model tiers or features emerge.
Time saved:15m 9s

Share this

Tags

Written by: 1 Minute Signal Editorial Team