Anthropic's Claude Opus 5.5 Launched with Improved Coding and Reduced Pricing
Anthropic, SpaceXAI, and Xiaomi released new models this week, intensifying competition as OpenAI prepares for its DevDay.
The story
Anthropic introduced Claude Opus 5.5 on September 22, showcasing enhanced coding capabilities and a reduced per-token price. This model achieved a 57.6 on the Artificial Analysis Intelligence Index v4.3.2, marking the highest score published to date and surpassing its predecessor, Claude Fable 5.1, by 4.3 points.
The pricing shift makes the new model more accessible at $4/$20 per million tokens, down from Claude Opus 5's $5/$25. This move comes as other labs also push new releases; SpaceXAI launched Grok 4.7 on September 21, boosting its coding and knowledge work performance, and Xiaomi debuted its open-source MiMo-V2.6-Pro with a 1.02-trillion-parameter mixture of experts, also on September 21. These releases highlight a rapid pace of innovation and competitive pricing adjustments across the AI software landscape.
Who moved
Anthropic
What Changed: Released Claude Opus 5.5 with improved coding capabilities and reduced pricing.
Consequence: This positions Opus 5.5 as a top-performing model while making it more cost-effective for users.
SpaceXAI
What Changed: Shipped Grok 4.7, a new base model with stronger performance in coding and knowledge tasks.
Consequence: Grok 4.7 improved its scores on various benchmarks, including CursorBench and DeepSWE.
Xiaomi
What Changed: Published MiMo-V2.6-Pro, a 1.02-trillion-parameter open-source model under an MIT license.
Consequence: The model achieved a high score on the Intelligence Index v4.3.2 and offers a low cost per index task.
China State TV Affiliate
What Changed: Accused Anthropic on September 19 of enabling U.S. intelligence access to user data through its privacy policy.
Consequence: This attack, reported on September 22, signals increased geopolitical scrutiny on Anthropic's data handling practices.
New models
Claude Opus 5.5
Lab: Anthropic
What: Offers enhanced coding capabilities and achieved the highest score on the Artificial Analysis Intelligence Index v4.3.2.
Use: Designed for complex coding and advanced tasks, now with a lower per-token cost of $4/$20 per million tokens.
Grok 4.7
Lab: SpaceXAI
What: A new and larger base model with improved performance for coding and knowledge work.
Use: Available on the API, Cursor, and Grok Build for users needing stronger agentic and long-task capabilities.
MiMo-V2.6-Pro
Lab: Xiaomi
What: A 1.02-trillion-parameter mixture of experts model with 42 billion active parameters and a 1M-token context.
Use: Released under an MIT license, it offers high performance on the Intelligence Index v4.3.2 at a low cost for open-source applications.
Market signals
AI and machine learning captured 77% of global venture deal value in the first half of 2026, up from 30% in 2023, according to HSBC.
Implication: This indicates a significant concentration of capital into a smaller number of larger AI funding rounds.
At Series D and beyond, the median AI company valuation was $4.25 billion, a 560% premium over non-AI counterparts valued at $644 million.
Implication: AI companies continue to command substantial premiums in later-stage funding rounds compared to other tech sectors.
Builders saying
Sam Altman, CEO of OpenAI. Hinted at major AI agent releases coming at the upcoming DevDay, suggesting potential disruption to the enterprise software market.
What we'll be watching
- OpenAI's DevDay conference on September 29 in San Francisco.
- OpenAI's keynote livestream for DevDay will be free and open to everyone.
- OpenAI plans DevDay Exchanges in eight global cities, including Bengaluru, Tokyo, and London.
- Potential for new AI agent releases at OpenAI DevDay, which could impact single-feature software tools.
Reporting + analyst voices: grounded via Google Search at publish time.