Anthropic’s AI assistant. Opus 4.8 is the flagship as of May 28, 2026, replacing Opus 4.7 for the hardest reasoning, coding, and agentic work while keeping standard Opus API pricing at $5 input / $25 output per million tokens. Sonnet 4.6 and Haiku 4.5 round out the lineup. The 1M token context window carries across Opus and Sonnet; Claude Code CLI handles agentic coding from the terminal; Constitutional AI training targets reduced sycophancy and harmful outputs. Anthropic’s May 28 Opus 4.8, dynamic workflows, and Series H announcements reinforce that Claude’s product story is now as much about execution controls, capacity, and enterprise trust as raw model quality.
Claude Design (launched April 17, 2026) is the prompt-to-prototype product in the Claude lineup: it produces slide decks, app mockups, marketing one-pagers, and pitch materials from text prompts. It reads a company’s codebase and design files to apply the existing design system automatically, then hands off to Claude Code for production builds. Exports to PDF, PPTX, URL, or Canva. Available to Claude Pro, Max, Team, and Enterprise subscribers. The launch sent Figma stock down ~7% on the day. See the dedicated tool page or the launch coverage.
Recent developments
- May 28: Claude Opus 4.8 launched with dynamic workflows, effort controls, Messages API instruction updates, and cheaper fast mode. This is now the model to evaluate for serious Claude reasoning, coding, and agentic work; dynamic workflows matter most for Claude Code buyers because they let Claude fan out large jobs across parallel subagents.
- May 28: Anthropic raised $65B at a $965B post-money valuation. Treat the round as a capacity and enterprise-scale signal, not a buying recommendation by itself: rate limits, latency, regional availability, support, and price discipline still need workflow-specific testing.
- May 25: Pope Leo XIV released an AI encyclical while Anthropic co-founder Chris Olah called for outside checks on AI labs. This is not a Claude feature launch, but it strengthens Anthropic’s public-trust positioning while keeping normal buyer diligence on pricing, limits, data handling, and implementation quality essential.
- May 22: Anthropic said Project Glasswing found 10,000+ high or critical vulnerabilities with Mythos Preview. The buyer signal is that Claude’s restricted cyber frontier is now shaping patch triage, disclosure capacity, and policy debate, not just benchmark charts.
- May 21: Claude added 28 compliance integrations while Opus security partners showed live cyber-defense use cases. The buyer signal is that Claude’s enterprise moat is shifting toward governed deployment, auditability, and specialist security channels.
- May 20: OpenAI IPO reporting, Anthropic profit forecasts, and SpaceX compute costs made frontier AI economics more visible. Claude buyers should track capacity, limits, and compute-contract exposure alongside model quality.
- May 19: Claude Managed Agents added self-hosted sandboxes and MCP tunnels. Enterprises can keep agent tool execution, files, packages, and private MCP servers inside customer-controlled infrastructure while Anthropic continues to run the agent loop.
- May 19: OpenAI co-founder Andrej Karpathy joined Anthropic’s pretraining team. It is not a product launch, but it is a meaningful research-talent signal behind future Claude model development.
- May 19: KPMG is making Claude available to more than 276,000 employees and embedding it inside Digital Gateway. The rollout strengthens Claude’s enterprise-distribution story through services firms that can implement governed client workflows.
- May 18: Anthropic acquired Stainless to strengthen Claude’s developer platform. Stainless already powered Anthropic’s official SDK generation; the acquisition points to more investment in SDKs, MCP tooling, CLIs, and agent connectivity.
- May 14: Anthropic expanded Claude through a PwC enterprise rollout and a $200M Gates Foundation partnership. PwC will roll out Claude Code and Cowork starting with U.S. teams and train 30,000 professionals; the Gates work funds Claude credits, grants, and technical support across health, life sciences, education, and economic mobility.
- May 14: Anthropic split Claude Agent SDK and
claude -pinto separate monthly credits. Starting June 15, programmatic usage leaves normal subscription limits and draws from credits worth $20 on Pro, $100 on Max 5x, and $200 on Max 20x before API-rate extra usage. Interactive Claude Code and Claude chat remain on normal subscription limits. - May 13: Anthropic launched Claude for Small Business with QuickBooks, PayPal, HubSpot, Canva, Docusign, Google Workspace, and Microsoft 365 connectors. It packages 15 ready-to-run agentic workflows across finance, operations, sales, marketing, HR, and customer service through Claude Cowork.
- May 14: Mythos, Claude Opus 4.7, and GPT-5.5-Cyber drove a “vulnpocalypse” of new CVEs in May. Palo Alto Networks found 75 holes / 26 CVEs in its own codebase with help from Mythos and Opus 4.7; Microsoft’s MDASH agentic system contributed to a record 30-critical-CVE Patch Tuesday; Mozilla shipped 423 Firefox bugs in April, 20x its prior monthly average. Project Glasswing members got patch lead time; the broader market has 3-5 months before parity.
- May 13: Anthropic is in talks for a $30-50B round at a $950B valuation, per the New York Times. The mark would put Anthropic ahead of OpenAI’s $825B and represent a 1,445% lift in 12 months. Dario Amodei told Code with Claude attendees the company has crossed a $30B annualized revenue run rate. Round not yet signed.
- May 13: Anthropic passed OpenAI in US business AI adoption for the first time per Ramp’s May 2026 AI Index. 34.4% of surveyed US businesses pay for Anthropic vs. 32.3% for OpenAI. Anthropic’s share went from ~9% in May 2025 to 35%+ today, a 26-point year-over-year jump driven by Claude Code and the Microsoft 365 Claude embedment.
- May 12: Anthropic launched Claude for Legal with 12 practice-area plugins, 20-plus MCP connectors (Harvey, Relativity, Everlaw, Thomson Reuters CoCounsel, DocuSign, Box), and a Microsoft 365 integration that embeds Claude across Word, Outlook, Excel, and PowerPoint. Freshfields, Quinn Emanuel, Holland & Knight, and Crosby joined as launch customers.
- May 11: Google said it caught the first AI-generated zero-day exploit before mass deployment. Google did not name the model used by the attacker. The buyer implication for Claude and Mythos is clear: defensive AI security claims now need explicit offensive-uplift, access-control, and disclosure evidence.
- May 10: Reports tied Anthropic to Akamai’s $1.8B frontier-model cloud commitment. The reporting reinforces that Claude capacity is becoming an infrastructure-market story, not just a model-quality story.
- May 9: FIS and Anthropic are developing a Claude-powered financial-crimes agent for banks. The project pushes Claude deeper into regulated, evidence-heavy enterprise workflows where auditability matters as much as reasoning quality.
- May 6: Claude Managed Agents added dreaming, outcomes, and multiagent orchestration. The release pushes Claude further into hosted agent infrastructure, with memory refinement, rubric-based outcomes, and parallel subagent work.
- May 6: Anthropic used SpaceX Colossus capacity to raise Claude Code and Opus API limits. Higher capacity reduces one of Claude’s biggest friction points for heavy agent and coding workloads.
- May 5: ServiceNow Action Fabric opened governed enterprise actions to Claude and other AI agents. Claude Cowork is an early design-partner surface for routing enterprise requests through ServiceNow approval chains.
- April 30: Claude Security entered public beta for Enterprise customers. Claude can now scan repositories for vulnerabilities, validate findings, and propose patches with Opus 4.7, strengthening Anthropic’s enterprise-security story beyond Mythos.
- April 30: OpenAI began rolling GPT-5.5 Cyber to critical defenders through Trusted Access. This narrows the contrast with Anthropic’s restricted Mythos access: both major labs are now gating advanced cyber-capable model access.
- April 30: Sources: Anthropic could raise $50B at $900B valuation, nearly 2.5x February’s $380B price. Revenue run rate now exceeds $30B, driven by Claude Code and agentic products. Would be one of the largest single AI financing rounds ever.
- April 29: Anthropic updated its Responsible Scaling Policy around external review. This matters for Claude because enterprise buyers now judge the assistant not only by model quality, but by how Anthropic handles capability thresholds, external scrutiny, and deployment controls.
- April 28: Claude for Creative Work adds connectors for Adobe, Blender, Ableton, Autodesk, and more. The update moves Claude deeper into design, 3D, audio, and live-visual production software through first-party connector surfaces.
- April 27: Adobe Firefly AI Assistant entered public beta, with Adobe saying it is also working to bring its pro-grade creative tools into third-party AI surfaces such as Claude.
- April 27: Cursor and Claude were named in a reported PocketOS database-deletion incident. The lesson is production guardrails: scoped credentials, immutable backups, and human approval for destructive actions.
- April 26: Anthropic presses agent-commerce controls in the press, keeping permissions, merchant boundaries, and user accountability central to Claude’s agent story.
- April 26: Anthropic outlines election-safeguard work, reinforcing safety and policy posture as part of Claude’s enterprise trust pitch.
- April 25: AI News Desk, April 25 grouped GPT-5.5 API, Copilot, Project Deal, and Google-Anthropic financing as the main frontier-lab weekend signals.
- April 24: Anthropic and NEC announce a 30,000-employee Claude rollout. NEC will make Claude available across its global workforce and package Claude into NEC BluStellar offers and cybersecurity services.
- April 24: Google reportedly plans up to $40B in cash and compute investment for Anthropic. If completed, it would further expand Claude’s multi-cloud capacity alongside AWS and Google TPU commitments.
- April 24: AI Industry Roundup, April 24 tracked Anthropic capital pressure; Project Deal showed a Claude-run marketplace experiment.
- April 23: AI Industry Roundup, April 23 compared GPT-5.5, Cloud Next, Grok, and EU AI funding signals around Claude’s competitive context.
- April 22: Anthropic commits multi-gigawatt capacity on next-gen Alphabet-Broadcom TPUs starting 2027. Extends the Ironwood 1M-TPU commitment into the 8th-generation silicon roadmap. Anthropic now has 10 GW+ of multi-cloud compute committed across AWS, Google, and Nvidia through 2030.
- April 22: Anthropic commits to up to 1M Google Ironwood TPUs as Google launches the 7th-gen inference-first TPU at Cloud Next 2026. Combined with the AWS pact, Anthropic now has 8.5 GW+ of multi-cloud compute committed for Claude.
- April 21: Kimi K2.6 Agent Swarm raised the open-weight agentic baseline; recursive-superintelligence safety analysis kept governance risk in view.
- April 21: Anthropic outspent OpenAI on federal lobbying in Q1 2026: $1.6M vs $1M. First quarter Anthropic has topped the frontier-lab Washington spend table.
- April 20: Amazon commits up to $25B more to Anthropic, tied to a $100B+ AWS spend pledge and up to 5 GW of dedicated compute for Claude training and serving. One of the largest AI infrastructure pacts ever disclosed.
- April 20: Simon Willison’s Claude system-prompt timeline made prompt-policy drift easier to inspect across Claude 3.0 through Opus 4.7.
- April 19: US courts split on AI chatbot privilege. Rakoff (SDNY) compelled production of 31 Claude-generated documents in US v. Heppner. Consumer Claude chats are discoverable by default. Enterprise tiers with zero-retention offer more protection.
- April 17: Claude Design launches. Prompt-to-prototype bundled with Pro+. Figma stock fell ~7%.
- April 17: Dario Amodei meets White House Chief of Staff over Mythos. Federal cybersecurity use of Anthropic’s restricted Mythos model on the table; first formal sit-down since the Pentagon’s “supply chain risk” designation. White House calls talks “productive and constructive.”
- April 17: Cursor, Windsurf, Zed, Continue, GitHub Copilot ship Opus 4.7 within 24 hours of the release.
- April 16: Opus 4.7 ships as new flagship, retaking frontier lead on agentic coding, tool use, and financial analysis.
- April 16: Systemic MCP vulnerability exposes 200k servers to arbitrary command execution. OX Security disclosure; Anthropic says behavior is by design, ships only documentation update advising STDIO adapter caution.
- April 16: Qwen3.6-35B-A3B released open-source. Aggregate ~82% of Opus 4.7 at zero cost; gap widens on MCP Atlas tool use (62%).
- April 14: Anthropic fielding $800B valuation offers, more than double February’s $350B round. Revenue at $30B annualised, 1,400% YoY. Early IPO talks with Goldman, JPM, Morgan Stanley for possible October 2026 listing.
- April 14: Claude Computer Use agent mode expands to broader availability.
- April 10: Mythos Preview access tightened over cybersecurity dual-use risk. Invitation-only via Project Glasswing.
- April 7: Anthropic locks in 3.5GW Google TPU capacity via Google Cloud + Broadcom. $30B run-rate disclosed.
System Verdict
Pick Claude if you need long-form reasoning, writing coherence, Claude Code, or a controlled enterprise workflow more than broad assistant coverage. Opus 4.8 (released May 28) is now Anthropic’s top model for complex reasoning, long-horizon agentic coding, and high-autonomy work. GPT-5.5 still makes ChatGPT the stronger default all-purpose product for many users, but Claude Code remains one of the strongest terminal coding agents on the market. The 1M context window on Opus and Sonnet is still a real advantage when you need whole-codebase or long-document work, and Anthropic is pushing harder than peers on explicit boundaries for agents that shop, transact, or interact with civic information.
Skip it if your workload needs image generation, video, or a deep plugin / integration ecosystem. Claude has no native image gen (use Midjourney or GPT Image 2), no video (Veo 3 or Kling), and a much smaller third-party tool marketplace than ChatGPT. Anthropic’s conservative safety posture also refuses more requests than ChatGPT does.
Who pays which tier: Free for casual Sonnet access, Pro $20/mo for most individuals needing Opus + extended thinking, Max 5x $100/mo for heavy solo coders, Max 20x $200/mo for teams or sustained agentic workloads. API developers should benchmark Opus 4.8 standard mode and fast mode on real prompts before moving budget-sensitive workloads.
Key Facts
| Flagship model | Claude Opus 4.8 (released May 28, 2026) |
| Other production models | Sonnet 4.6 (speed/quality balance) · Haiku 4.5 (fast, high-volume) |
| Claude Design | Prompt-to-prototype tool · launched April 17, 2026 · research preview · Pro/Max/Team/Enterprise only |
| Context window | 1M tokens on Opus and Sonnet · 200K on Haiku |
| Max output | 128K (Opus) · 64K (Sonnet) · 64K (Haiku) |
| API pricing | Opus 4.8: $5 in / $25 out per MTok standard; $10 / $50 per MTok fast mode |
| Sonnet 4.6: $3 / $15 per MTok · Haiku 4.5: $1 / $5 per MTok | |
| Subscription pricing | Free · Pro $20 monthly or $17/mo annual · Max 5x $100 · Max 20x $200 · Team $20/seat annual ($25 monthly) standard or $100/seat annual ($125 monthly) premium · Enterprise $20/seat plus API-rate usage |
| Agentic coding | Claude Code CLI · top terminal coding agent; GPT-5.5 Codex is now the main overall challenger |
| Enterprise distribution | NEC partnership announced April 24, 2026 · about 30,000 NEC employees plus BluStellar and cybersecurity services |
| Recent infrastructure financing | Anthropic raised $65B at a $965B post-money valuation on May 28, 2026, citing Claude demand, compute expansion, and enterprise scaling |
| Design tooling | Claude Design · prompt-to-prototype · exports PDF, PPTX, URL, Canva · hands off to Claude Code for build |
| Image generation | None (use Midjourney or GPT Image 2 alongside) |
| Video generation | None (use Veo 3 or Kling alongside) |
| Separate frontier model | Mythos Preview (cybersecurity, invitation-only via Project Glasswing) |
| Trust posture | Agent-commerce controls, election safeguards, no-training-on-data team tiers, and conservative refusal policy are core buying criteria |
What it actually is
A single consumer + API product covering text chat, long-form analysis, code, and agentic coding via Claude Code CLI. All current models support extended (visible chain-of-thought) thinking.
Prompt caching drops cache-hit input costs to 10% of standard. The Batch API cuts both input and output by 50%, materially cheaper than OpenAI for bulk workloads.
The real moats: the 1M token context at flat per-token rates, Constitutional AI training that produces more structurally coherent output on long documents, and Claude Code CLI, which autonomously reads and writes across full codebases from the terminal. After the May 28 update, dynamic workflows make Claude Code more credible for broad jobs that benefit from parallel subagents and independent verification.
When to pick Claude
- Long-form writing or document analysis. Better coherence on many 10K+ word outputs than ChatGPT; the 1M context means book-length inputs load in one session without chunking.
- Agentic coding from the terminal. Claude Code is the strongest CLI agent; Ultraplan auto-creates cloud environments, the Monitor tool watches running processes.
- Legal, compliance, civic, or regulated workflows. Constitutional AI training reduces sycophantic agreement with user-stated premises, and Anthropic is explicitly publishing safeguards around elections and transactional agents. That’s the exact failure mode that can kill premise-testing tasks on more permissive assistants.
- Security research via Mythos Preview. Scores 83.1% on vulnerability-reproduction benchmarks (vs Opus 4.6’s 66.6%). Invitation-only via Project Glasswing consortium.
- API bulk workloads. Batch API + prompt caching combine for genuinely cheaper Opus runs than some equivalent GPT-5.5 Pro workloads.
When to pick something else
- Image generation: Midjourney (quality) or ChatGPT’s GPT Image 2 (bundled with text chat). Claude has none.
- Video generation: Veo 3 (via Gemini) or Kling. Claude has none.
- Google Workspace integration: Gemini 3.1 Pro. Claude’s Workspace hooks are thin.
- Broad plugin / custom-agent marketplace: ChatGPT. GPT Store has no Claude equivalent.
- One tool for everything (casual use): ChatGPT. Claude is a specialist, not a generalist.
Pricing
Subscription pricing via claude.com/pricing:
| Plan | Price | Models | Who’s it for |
|---|---|---|---|
| Free | $0 | Sonnet (capped daily messages) | Casual use, no extended thinking |
| Pro | $20/mo or $17/mo annual | Opus 4.8 · Sonnet 4.6 · Haiku 4.5 · extended thinking · Projects | Most individuals should land here |
| Max 5x | $100/mo | Full lineup, 5× Pro usage | Heavy solo coders / writers |
| Max 20x | $200/mo | Full lineup, 20× Pro usage | Teams or sustained agentic workloads |
| Team (Standard seat) | $20/seat annual or $25/seat monthly | Full lineup + admin + no-training-on-data; 5-150 seats | Small/mid teams |
| Team (Premium seat) | $100/seat annual or $125/seat monthly | Higher usage, priority access | Heavier per-seat workloads |
| Enterprise | $20/seat + API-rate usage | Full lineup + SSO + SCIM + audit + Compliance API + HIPAA-ready | Compliance-heavy orgs |
API pricing via platform.claude.com:
| Model | Input ($/MTok) | Output ($/MTok) | Context | Max Output |
|---|---|---|---|---|
| Opus 4.8 | $5 | $25 | 1M | 128K |
| Sonnet 4.6 | $3 | $15 | 1M | 64K |
| Haiku 4.5 | $1 | $5 | 200K | 64K |
Fast mode on Opus 4.8 is $10 input / $50 output per MTok. Batch API discounts all models 50%. Prompt caching drops cache-hit input to 10% of standard.
Prices verified 2026-05-29 via Claude pricing, Anthropic model docs, and Anthropic’s Opus 4.8 launch.
Opus 4.8 cost-impact note: standard pricing matches Opus 4.7, while fast mode is now $10 / $50 per MTok. Budget-sensitive API workloads should benchmark standard versus fast mode and set effort levels explicitly before migrating.
Against the alternatives
| Claude Opus 4.8 | GPT-5.5 Pro | Gemini 3.1 Pro | |
|---|---|---|---|
| Agentic coding | Claude Code (strongest CLI) | Codex (strongest IDE integration) | No native agent |
| Image generation | None | GPT Image 2 | Imagen 4 (built-in) |
| Video generation | None | None | Veo 3 (built-in) |
| Long-context | 1M tokens, flat rate | Undisclosed window | 1M tokens |
| Workspace / ecosystem | Narrower | Largest (GPT Store) | Best for Google |
| Default reasoning quality | Strongest on long-form | Broadest feature suite | Strongest Google-stack |
| Best viewed as | Reasoning + writing + trust specialist | Generalist default after GPT-5.5 | Google-stack integrator |
Failure modes
- Copilot-hosted Claude model availability changes fast. GitHub retired Claude Sonnet 4 in Copilot and points users to Sonnet 4.6; Rubber Duck in Copilot CLI can now use Claude as a cross-model critic. Claude access inside third-party tools should be verified on the exact platform, not assumed from claude.ai.
- Conservative safety refusals. Claude declines requests that are not actually harmful at a higher rate than ChatGPT. Noticeable on red-team research, edgy creative writing, and some analysis of adversarial content.
- No native image or video. Workflows that need multimodal output need a second tool alongside Claude.
- Opus 4.8 usage controls need testing. Same standard sticker price as Opus 4.7 and cheaper fast mode, but effort settings and dynamic workflows can change real usage quickly. API and Claude Code teams need prompt-level benchmarks, usage alerts, and policy gates before broad migration.
- Computer Use is still a research preview. Available inside Claude Code only, not in claude.ai chat. Not production-ready.
- Mythos Preview is invitation-only. Advertised in Project Glasswing materials but most users cannot access it.
- Memory is session-scoped. Cross-conversation persistent memory is available through Projects with explicit context files, but there is no implicit cross-session memory like some ChatGPT configurations.
- Rate limits unpublished. Pro / Max tier message caps are not publicly specified and vary with demand. Heavy users discover limits by hitting them.
- Smaller plugin / agent ecosystem. No equivalent to the GPT Store. Integrations ship through the Anthropic-maintained MCP registry instead, which has narrower breadth.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility × Value × Moat × Longevity, unweighted average). Last verified 2026-05-29 against Claude pricing, Anthropic’s model docs, the Opus 4.8 launch, the Series H funding update, the May 25 Vatican AI encyclical coverage, the May 21 Claude compliance and security-partner update, the May 20 frontier AI capital-pressure update, the May 19 KPMG rollout, the May 18 Stainless acquisition, the May 14 PwC/Gates expansion, the May 14 Agent SDK credit split, the May 12 Claude for Legal launch, the May 10 reported Akamai cloud-deal coverage, the May 9 FIS financial-crimes agent coverage, the May 6 Managed Agents update, the May 7 GitHub Copilot Claude model notices, the May 7 Copilot Rubber Duck update, and the NEC partnership.
FAQ
Is Claude free to use?
Yes. The free tier gives capped daily access to Sonnet. Claude Pro at $20/month unlocks Opus 4.8, extended thinking, Projects, and higher message limits.
What changed in Opus 4.8?
Opus 4.8 replaced Opus 4.7 as Anthropic’s most capable model on May 28, 2026. Standard API pricing stayed at $5 input / $25 output per million tokens, fast mode dropped to $10 / $50 per million tokens, and the release added effort controls, Messages API instruction updates, and Claude Code dynamic workflows.
What is Claude Code?
A CLI coding agent that reads, writes, and manages full codebases from the terminal. Supports Ultraplan (cloud environment creation, early preview), the Monitor tool for watching running processes, and a Computer Use capability in research preview.
What is Claude Design?
A new (April 17, 2026) prompt-to-prototype product bundled with Claude Pro and higher tiers. Turns text prompts into slide decks, app mockups, one-pagers, and pitch materials. Reads your codebase and design files to extract the design system automatically, so outputs stay on-brand. Exports to PDF, PPTX, URL, or Canva. Hands off to Claude Code for production builds. Direct competition to Figma, Lovable, Bolt.new, and v0. Figma stock fell ~7% on launch. Currently research preview; not in the free tier. See the full Claude Design tool page.
What is Claude Mythos Preview?
A separate frontier model specialized for defensive cybersecurity, announced as part of Project Glasswing (a consortium including AWS, Apple, Google, and Microsoft). Not a general-release product. Access requires an invitation through the Glasswing program.
How does the 1M context compare?
Opus 4.8 and Sonnet 4.6 both support 1M tokens at flat per-token pricing (no long-context surcharge). Gemini 3.1 Pro also offers 1M tokens. ChatGPT plan context windows differ from OpenAI API context windows, so compare the exact surface before buying. At equivalent context lengths, Claude maintains strong coherence in published third-party evaluations.
Related
- Category: AI Chatbots · AI Coding · AI Writing
