WSJ: Anthropic draft IPO filing shows red-hot growth, risks, and massive computing costs

Claude is Anthropic's AI assistant based on the Constitutional AI and RLHF-aligned methodology. Launched in 2023, Claude emphasizes helpful, harmless, and honest interactions. It supports extremely long context windows of up to 200K tokens, making it effective for analyzing lengthy documents. Claude 3.5 Sonnet represents the mid-tier model with strong coding and reasoning capabilities. The iOS app and web interface provide easy access across devices.

Cerebras documents Shared Inference and Dedicated Inference at inference-docs.cerebras.ai. Shared Inference on Free Trial and Pay as You Go tiers lists models such as gpt-oss-120b (about 3000 tokens/s, 65k context on free / 131k on paid) and qwen-3.8-27b (about 1850 tokens/s, 64k / 128k) per the model catalog. Chat completions use the Cerebras Cloud SDK (`cerebras.cloud.sdk` in Python, `cerebras_cloud_sdk` in TypeScript) with `client.chat.completions.create()`, message roles including system, user, assistant, tool, and developer, plus optional `reasoning_effort` and `stream`. Docs describe tool calling (`tool_choice` none/auto/required or a named function, strict mode, parallel calls) for gpt-oss-120b and qwen-3.8-27b on Shared Inference. Additional model families, reserved capacity, and production SLAs are documented under Dedicated Inference. The catalog states Shared Inference models are unpruned originals, with selective weight-only quantization for storage and full-precision activations.

OpenRouter is a model gateway that exposes many third-party AI models through one OpenAI-compatible API. Teams can compare providers, set routing preferences, and switch models without rewriting core client logic for each vendor SDK. The service publishes per-model pricing and supports pay-as-you-go usage.

Anthropic documents remote Model Context Protocol connectors in Claude Help Center and platform.claude.com docs: users add publicly reachable MCP server URLs so Claude clients (claude.ai, Claude Desktop, Cowork, mobile) connect from Anthropic cloud infrastructure rather than the local machine. Team and Enterprise admins add connectors in Admin settings; individuals enable them under Settings > Connectors. The Messages API MCP connector (beta header anthropic-beta: mcp-client-2025-11-20) accepts mcp_servers entries with type url, name, and https URL, paired with mcp_toolset tools entries; servers must support streamable HTTP/SSE and be internet-accessible to Anthropic IP ranges.

Airtable documents an official remote MCP server at https://mcp.airtable.com/mcp (Streamable HTTP) in developer docs and the Help Center. It is available on all plans. Tools cover searching and listing authorized bases, creating tables and fields, reading and writing records and comments, building interfaces, and managing automations (`list_automations`, `get_automation`, `create_automation`, `get_create_automation_instructions`). Docs say `create_automation` saves a draft that stays off until a user turns it on in the Airtable UI. Authentication is OAuth (recommended, including Dynamic Client Registration) or a personal access token with scopes such as data.records:read/write, schema.bases:read/write, data.recordComments:read/write, and workspacesAndBases:read. Permissions mirror Airtable roles: Owners and Creators can change schema and records; Editors can change records; interface-only users see interface pages. The Help Center states MCP calls follow standard Airtable API rate limits and that creating records is limited to 10 records per request. Omitting write scopes yields read-only agent access.

Structures Wall Street Journal reporting on October 20, 2026 that the Journal viewed a new draft prospectus for Anthropic's planned IPO into a finance, customer-concentration, and compute-commitment checklist. The workflow separates verified facts—the draft shows rapid revenue growth with many risks and large computing costs; first-half 2026 net loss of $80.7 billion, which WSJ reporter Anissa Gardizy tied to accounting adjustments, versus a $13 billion net loss in the same period a year earlier; first-half 2026 revenue of $16.3 billion versus $1.1 billion in the first half of 2025; about 20% of revenue from two customers; Anthropic could spend $750 billion in coming years to obtain more computing capacity, including agreements still under negotiation; risks include a narrow customer base and uncertain access to needed compute—from internal vendor, treasury, and model-access decisions. Distinct from earlier confidential-filing headlines that did not include these first-half figures.

Structures CNN reporting on June 30, 2026 that the US government lifted export controls on Anthropic's Claude Fable 5 and Mythos 5 into a policy, security, and release-governance checklist. The workflow separates verified facts—Anthropic said Commerce lifted export controls on Fable 5 and Mythos 5 and would begin restoring access; Commerce Secretary Howard Lutnick posted on X about lifting controls on Fable after two weeks working with Anthropic to analyze and approve Fable 5; Fable is Mythos with extra public guardrails; earlier export ban followed Amazon finding a jailbreak (Anthropic called jailbreaks simple and noted similar work-arounds on other public models); Anthropic implemented a new safeguard blocking reported vulnerabilities; Commerce had required suspending all foreign-national use including Anthropic employees; Mythos was later released to select government-approved entities; CNN notes experts say Mythos can exploit cyber vulnerabilities at unprecedented pace; White House also asked OpenAI to limit GPT 5.6 release to government-approved partners—from internal frontier-model access planning. Distinct from June ban-imposition reporting tracked by anthropic-mythos-export-control-due-diligence.