`DeepSeek V4 Pro` makes its 75% API discount permanent

// related

• #0001

  #0001Models & APIAnthropic TypeScript SDK4 hours ago

  `anthropic-sdk-typescript` `0.99.0` adds custom file size caps

  40radar

  File upload limits can now be tuned per use case. The streaming fix is small but useful for agents that depend on exact stop metadata.

  • —Custom file size caps landed via #1029; upload-heavy apps get more explicit guardrails before hitting provider limits.
  • —Streaming now carries stop_details through message_delta accumulation, reducing lost termination context in long responses.
  • —This is a maintenance release, not a product shift. Upgrade when file uploads or stream-stop handling are in your path.

  Source: github.com/anthropics/anthropic-sdk-typescript/releases/Read original →
• #0002

  #0002Models & APIAnthropic TypeScript SDK6 hours ago

  `Anthropic TypeScript SDK` `0.98.1` fixes skill upload paths

  40radar

  Anthropic TypeScript SDKLLM API SDK — call Anthropic APIs from TypeScript

  A narrow patch keeps directory prefixes when uploading skills.versions.create files. Upgrade only if your workflow packages nested skill assets.

  • —skills.versions.create now preserves directory prefixes during uploads, fixing broken layouts for nested skill files.
  • —The release also moves the client package to Trusted Publishing, reducing release credential risk rather than changing runtime behavior.
  • —The managed-agents private sandbox worker example was renamed to self-hosted-sandbox-worker; docs/scripts may need a path update.

  Source: github.com/anthropics/anthropic-sdk-typescript/releases/Read original →
• #0003

  #0003Models & APIr/OpenAI3 days ago

  DeepSeek launches API at 1/30th the price of GPT-5.5

  100radar

  DeepSeekHigh-performance LLM API — Radically cheaper than OpenAI/Claude

  The era of high-margin AI APIs may be over. DeepSeek offers a 'good enough' alternative at a fraction of the price, shifting the competition to cost-effectiveness and altering the economics of building AI features.

  • —DeepSeek is 11.5x cheaper on input and 34.5x cheaper on output compared to GPT-5.5, and 28.7x cheaper than Claude Opus on output.
  • —As model performance begins to plateau across the board, price is emerging as the primary competitive axis. This may signal the beginning of the 'good enough' model era.
  • —This makes high-volume use cases, previously blocked by cost, a realistic option for indie products, such as large-scale data processing or frequent API calls.
  • —The market's reaction suggests that AI is not a winner-take-all market, but one where margins will compress as viable, cheaper alternatives emerge.

  Source: www.reddit.com/r/OpenAI/comments/1tm49d0/deepseek_just_pRead original →