The max request size for Messages and Token Counting on the Claude API has quadrupled from 32 MB to 128 MB. Long computer-use sessions and large PDFs were frequently hitting the old limit before exhausting the context window.
The 5-hour session limit is being lifted for verified Ultra accounts, part of the broader Max plan rename (Max 5x to Max, Max 20x to Ultra).
Users can confirm verification status in Settings > Usage: a missing "Current session" bar means the account is verified. The 5-hour limit is removed for verified Ultra accounts.
Verification is determined automatically on Anthropic's end for trusted accounts, with details available at a linked support page.
For requests with more than 20 images, the per-image pixel limit has been raised from 2000 px to 3000 px. The 128 MB cap applies on the first-party Claude API; cloud platform limits remain between 20 and 32 MB.
There's a real pattern where heavy coding-agent use creates a feeling of productivity and constant work while actual output stalls. Thariq thinks this mode is often a form of avoidance rather than genuine ambition.
Thariq (former Claude Code team) pushes back on the framing that coding agents are harmful, clarifying the point is narrower: the all-night, revelation-chasing mode of using them can be a sign of avoidance, not that agents themselves aren't genuinely useful.
A founder described spending 14 months building infrastructure with agents before refocusing on real business outcomes, and Thariq called that arc a common and ultimately healthy trajectory.
The false-confidence productivity loop is mostly a solo or small-startup phenomenon; enterprise users who struggle with AI productivity usually haven't updated their planning cadence rather than falling into the mania pattern.
Responding to a critique that coding agents create a goldfish-like loop because they don't help users track progress, Thariq agreed the products could do better at continuity and memory across a session.
