Tech • AI • Robotics • Game
Huge AI News: GPT-6 Sol & Luna, Opus 5.5, Sonnet 5.5, Haiku 5.5, Qwen 4.0, & Trump to Change AI!
9.5/10TL;DR
OpenAI, Anthropic, Alibaba, Tencent, and Donald Trump all drove a turbulent AI news cycle, with cheaper high-performance models, new benchmark claims, upcoming releases, and a proposed political rebranding of AI as “super intelligence.”
KEY POINTS
OpenAI launches GPT6 Soul and Luna
OpenAI introduced GPT6 Soul and GPT6 Luna as lower-cost models intended to bring near-flagship capability to routine workloads. The company positioned them as upgrades in coding, factuality, computer use, professional tasks, and alignment while cutting API prices by 50% from GPT 5.6 promotional levels.
Sharp API price cuts
GPT6 Soul is priced at $2 per 1 million input tokens and $10 per 1 million output tokens. GPT6 Luna drops to $0.10 per 1 million input tokens and $0.50 per 1 million output tokens, a level that significantly lowers the cost of deploying advanced models at scale.
Business workflow performance improves
In Automation Bench, which measures real business workflows across apps, GPT6 Soul at high effort reportedly outperformed Claude Opus 5 at max effort while costing only 9% as much per task. Luna also improved, gaining 5.44 percentage points over its predecessor at high effort while reducing per-task cost by 58%.
Benchmarks show efficiency more than a large capability jump
On broader rankings, GPT6 Soul placed just behind GPT6 Astra and was described as third overall on one platform. Separate testing indicated only about a one-point gain over the prior GPT 5.6 Soul, suggesting the biggest story is not a dramatic leap in intelligence but better speed, pricing, and output quality per dollar.
Hallucination rates fall sharply
One of the clearest reported improvements was factual reliability. Soul saw its hallucination rate fall from 92% to 60%, while Luna dropped from 93% to 77%, a notable reduction even if those rates remain high in absolute terms.
Anthropic’s Claude Opus 5.5 sets the pace
Anthropic released Claude Opus 5.5, which was described as the top-ranked model currently available. It reportedly performs around the level of Claude Fable 5.1 across many tasks while costing 40% less to run than Opus 5, reinforcing an industry trend toward lower-cost frontier performance rather than pure capability races alone.
Opus 5.5 shows stronger high-end coding and 3D generation
In side-by-side coding and Three.js style tests, Opus 5.5 produced more polished and interactive scenes than GPT6 Soul, including richer shaders, textures, lighting, and proper render loops. In one four-scene comparison, Opus 5.5 generated higher-detail outputs but cost about $4.37, versus roughly $0.34 for GPT6 Soul, making the trade-off between quality and price unusually stark.
Game and simulation demos highlight a widening top-end gap
Advanced demos attributed to Opus 5.5 included a night-train simulation, a shader-heavy sandbox game resembling Minecraft, a Spider-Man style project, and an Elden Ring inspired action game with a boss fight. One reported build ran for 1 hour and 37 minutes at max effort, underscoring the model’s ability to sustain long, detailed coding sessions.
More Claude 5.5 models are expected soon
Anthropic also signaled that Claude Sonnet 5.5 and Claude Haiku 5.5 are expected in the coming weeks. That suggests the company may mirror OpenAI’s strategy by extending improved efficiency, safety, and capability down the model stack.
Alibaba and Tencent signal more competition
Alibaba is preparing a broader Qwen 4 family, including Qwen 4 Max, Qwen 4 Plus, Qwen 4 Flash, and Qwen 4 27B. Reports also point to future Qwen 4.5 and Qwen 5 systems that could scale to 5 trillion to 10 trillion parameters. Meanwhile, Tencent launched HY Image 3.5 Preview, claiming a 30% higher win rate in human evaluations than HY Image 3.0, support for up to 2K generation, and pricing of $0.02 per image.
Trump proposes renaming AI
Donald Trump argued that “artificial intelligence” makes the technology sound fake and pushed alternatives such as super intelligence and superior intelligence. In a public poll, superior intelligence won with 52% of roughly 184,000 votes, and he later said the US government would use “super intelligence” or “SI” in official documents.
CONCLUSION
The latest wave of releases shows the AI market splitting into two races at once: one for the absolute best model and another for the best capability per dollar. The biggest immediate shift may be economic, as lower pricing and stronger mid-tier systems make advanced AI much more practical for mainstream use.
Ask a question