Claude, Catch Me Up: What Happened While You Were Away
If you last used Claude a while ago and everything looks renamed, this page is the fast way back in. No launch-day hype — just what shipped, when, and…
Claude guide · as of July 10, 2026 · 4 minutes read · details change — confirm current specs on claude.ai
If you last used Claude a while ago and everything looks renamed, this page is the fast way back in. No launch-day hype — just what shipped, when, and whether it matters to you.
The big picture
Claude is now in its Claude 5 generation. Two models are current, and they sit at different tiers:
- Opus 4.8 (2026-05-28) is the current flagship Opus — the reliability-focused workhorse. Its headline is trust: it is roughly 4x less likely than Opus 4.7 to let its own code flaws slip through. It runs a 1M context window at $5/$25 per million tokens.
- Fable 5 (2026-06-09) is a tier above Opus — the first publicly available model built on the Mythos-class weights. It posts the highest benchmarks of any generally available model, runs 1M context, and costs $10/$50.
The other thing worth knowing is the pricing and context inflection that happened across the Opus 4.5 → 4.6 stretch. Opus 4.5 (2025-11-24) reset Opus pricing to $5/$25 with far fewer output tokens per answer, and Opus 4.6 (2026-02-05) brought the 1M context window to general availability with no surcharge. That combination is why current Opus is both cheaper and far larger than the Claude 4 models it replaced.
The timeline
Newest first. One honest line each.
| Date | Model / event | What it means |
|---|---|---|
| 2026-06-09 | Claude Fable 5 | First public Mythos-class model, a tier above Opus; 1M context, $10/$50, highest GA benchmarks; refusals return stop_reason "refusal"; 30-day retention is mandatory. |
| 2026-06-09 | Claude Mythos | The Mythos-class weights (above the Opus tier) that power Fable 5. |
| 2026-05-28 | Claude Opus 4.8 | Current flagship Opus; reliability-focused (~4x less likely than 4.7 to pass its own code flaws); 1M context; $5/$25; effort levels through xhigh/max; Fast mode at $10/$50. |
| 2026-04-16 | Claude Opus 4.7 | Tripled image resolution to 3.75MP; added "xhigh" effort and self-verification; 87.6% SWE-bench Verified; $5/$25; still active, no longer the flagship. |
| 2026-02-17 | Claude Sonnet 4.6 | The default Sonnet; 1M context. |
| 2026-02-05 | Claude Opus 4.6 | Brought 1M context to general availability with no surcharge; 78.3% MRCR v2 at 1M; 14.5-hour task window; native Agent Teams; $5/$25. Superseded by 4.8. |
| 2025-11-24 | Claude Opus 4.5 | The efficiency inflection: price reset to $5/$25 with ~76% fewer output tokens; 200K context (1M in beta). Still active as the affordable 200K Opus fallback. |
| 2025-10-01 | Claude Haiku 4.5 | Fast tier; active; 200K context, knowledge cutoff February 2025. |
| 2025-09-29 | Claude Sonnet 4.5 | Superseded by later Sonnets. |
| 2025-05-22 | Claude Opus 4 | Claude 4 flagship, "world's best coding model" at launch (72.5% SWE-bench Verified); $15/$75. Now retired. |
| 2025-05-22 | Claude Sonnet 4 | Claude 4 balanced tier; launched top of SWE-bench Verified at 72.7%; $3/$15. Now retired. |
| 2025-02-24 | Claude 3.7 Sonnet | Retired. |
| 2024-06-20 | Claude 3.5 Sonnet | Retired. |
| 2023-03-14 | Claude Instant | The original fast tier; retired. |
What actually changed for you
A few shifts matter more than the version numbers:
- Pricing reset downward. Opus went from the Claude 4 era's $15/$75 to $5/$25 at Opus 4.5, and it has held there through 4.8. Opus 4.5 also generated far fewer output tokens per answer (~76% fewer), so the same task typically costs less than the sticker change alone suggests. Fable 5 sits above that at $10/$50.
- 1M context is now standard, not a surcharge. Opus 4.6 made the 1M-token window generally available at no extra cost, and it carries through Opus 4.7, Opus 4.8, Sonnet 4.6, and Fable 5. You can hand these models much larger inputs than the old 200K ceiling allowed. Opus 4.5 and Haiku 4.5 stay at 200K.
- Effort levels are a real dial. Opus 4.7 introduced an "xhigh" effort setting plus self-verification, and Opus 4.8 exposes effort levels through xhigh/max. Higher effort means more internal checking before an answer lands — useful for code and analysis, slower for quick chores.
- These are cloud-only, and they have a knowledge cutoff. Every model here runs on Anthropic's servers via claude.ai and anthropic.com — none run on your own machine. Each also has a training cutoff (Haiku 4.5's is February 2025, for example), so for anything recent you still need to give it current sources rather than trust its memory.
One honest caveat on Fable 5: it enforces a mandatory 30-day retention window, and its refusals come back with stop_reason set to "refusal" — worth knowing if you are building on the API rather than chatting.
Related
- Flagship — which model is the current top pick, and when to reach past it.
- Compare the models — Opus 4.8 vs Fable 5 vs the fast and fallback tiers, side by side.
- Plans and pricing — the $5/$25 and $10/$50 numbers in context.
- Version history — the full lineage, retired models included.
