In partnership with
Hey there! 👋
Welcome back to SavvyMonk, your one-stop for AI and tech news that actually matters.
Anthropic has shipped Claude Sonnet 5, its most agentic Sonnet model to date, and priced it below every comparable model from OpenAI and Google. The launch lands weeks after a confidential IPO filing that could value the company near one trillion dollars.
Let's get into it.
The best prompt engineers aren't typing. They're talking.
Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.
Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.
89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.
TODAY'S DEEP DIVE
Claude Sonnet 5 Launches at Half the Price of Opus 4.8, With Performance Close Behind
Anthropic released Claude Sonnet 5 on 30 June 2026, calling it the most agentic Sonnet model the company has built. It is now the default model for Free and Pro plan users and is available across Max, Team, and Enterprise plans, as well as in Claude Code and on the Claude Platform.
Developers can call it through the API using the model name claude-sonnet-5, and Anthropic raised rate limits across Chat, Cowork, Claude Code, and the Claude Platform to handle the heavier token usage that comes with running the model at higher effort levels.
Closing the Gap With Opus
The organization claims is that Sonnet 5 performs close to Opus 4.8 while costing far less to run. On one agentic coding benchmark, Sonnet 5 scored 63.2 percent, ahead of Sonnet 4.6's 58.1 percent and trailing Opus 4.8's 69.2 percent, and on at least one knowledge work benchmark Sonnet 5 edged slightly past Opus 4.8 entirely.
Anthropic attributes the jump to gains in reasoning, tool use, software coding, and general knowledge work rather than any single new capability, and early access partners including Cursor and Zapier described the model finishing multi-step jobs end to end where earlier Sonnet versions used to stall partway through.
The company also tested Sonnet 5 against Opus 4.8 on agentic search and computer use tasks across a range of effort levels, since Anthropic now lets developers dial reasoning effort up or down depending on how much accuracy a task needs. Across those tests, Sonnet 5 narrowed the cost performance gap that previously separated it from Opus class models, giving developers a genuine middle option rather than a binary choice between a cheap model that struggles with longer tasks and an expensive one that handles them comfortably.
Cheaper Than the Competition
Pricing is where Anthropic is making its sharpest pitch. Sonnet 5 launches at an introductory rate of two dollars per million input tokens and ten dollars per million output tokens, holding through 31 August 2026, before rising to three dollars and fifteen dollars respectively.
That sits well below Opus 4.8's five dollars and twenty five dollars, and undercuts current pricing for OpenAI's GPT-5.5 and Google's Gemini 3.1 Pro, though it still costs more than Google's lighter Gemini 3.5 Flash. Anthropic also introduced a new tokenizer with this release, which means the same input can now map to roughly 1.0 to 1.35 times more tokens than before, and the company set its introductory pricing specifically to keep the transition close to cost neutral for existing users.
The Safety Picture
Anthropic's own evaluations found Sonnet 5 hallucinates and shows sycophantic behaviour less often than Sonnet 4.6, and that it is better at refusing malicious requests and resisting hijack attempts during prompt injection attacks.
On the company's broader behavioural audit, which checks for cooperation with misuse and other misaligned behaviour, Sonnet 5 scored lower, meaning safer, than Sonnet 4.6, though it still trailed both Opus 4.8 and the unreleased Claude Mythos Preview.
Rates of misaligned behavior on Anthropic’s automated behavioral audit, which tests for a very wide range of undesirable behaviors across many situations and contexts. Sonnet 5 shows an overall lower rate of misaligned behavior than Sonnet 4.6
On cybersecurity specifically, Anthropic said it did not deliberately train Sonnet 5 on cyber tasks, and in testing built with Mozilla around vulnerabilities in Firefox 147, neither Sonnet model produced a working exploit. Sonnet 5 showed a partial success rate of 13.2 percent against Sonnet 4.6's 8.8 percent, both well behind Opus 4.8's 68.8 percent and Mythos 5's 88.4 percent. Anthropic has kept real time cyber safeguards switched on by default for Sonnet 5 regardless, matching the protections already running on Opus 4.7 and 4.8.
The Shadow of Fable and Mythos
The release also lands against an awkward backdrop. Earlier in June 2026, the Trump administration directed Anthropic to restrict access to its most capable models, Fable 5 and Mythos 5, for foreign nationals both inside and outside the United States, citing concern that the models could be misused for hacking, and Anthropic pulled both offline as a result.
Sonnet 5 carries none of those restrictions and shipped without interruption, leaving Anthropic's mid-tier model as the most capable Claude version currently available to most of the world while its flagship models sit in limbo.
That contrast shapes how Anthropic is talking about Sonnet 5's own cyber capabilities. The company has been explicit that Sonnet 5 was never trained for cybersecurity work and trails Opus 4.8 and Mythos 5 by a wide margin on every relevant test, language that reads less like a limitation and more like reassurance aimed at regulators watching the export situation closely. A model that is good at almost everything except the one category currently under government scrutiny is, for now, the safer model to keep shipping freely.
The IPO Backdrop
Sonnet 5 also arrives as Anthropic pushes toward a public listing. The company confidentially filed paperwork with the Securities and Exchange Commission on 1 June 2026, with reporting pointing to a valuation approaching one trillion dollars, and OpenAI filed its own confidential paperwork roughly a week later.
Pricing Sonnet 5 below its competitors' equivalent tiers gives Anthropic a visible, easy to repeat selling point heading into that process, namely that enterprises can get near flagship performance without flagship bills.
The Bottom Line
Sonnet 5 is Anthropic's clearest statement yet that the model race has shifted from raw benchmark wins to cost per task. It is not the company's most capable model, and it is not trying to be.
It is the model Anthropic can point to while its actual flagships are tied up in export restrictions and its IPO paperwork sits with regulators, and on price alone it is hard for budget conscious developers to ignore.
AI PROMPT OF THE DAY
Category: Vendor Evaluation
"Act as a procurement analyst helping me decide whether to migrate a workload from one AI model to a cheaper alternative. Ask me about my current model, my monthly token spend, the task type like [task type], and how much accuracy loss I can tolerate. Then give me a structured framework for testing the cheaper model against my real workload before committing, including which metrics to track and how long the trial should run."
ONE LAST THING
A year ago, every model launch was a benchmark chase. Sonnet 5 is Anthropic admitting, in pricing if not in words, that most developers do not need the very best model, they need a good enough model that does not stall halfway through a task and does not wreck the monthly bill. That is a less exciting story than a new leaderboard topper, but it is probably the more honest one about where AI adoption actually happens.
Hit reply, I read every response.
See you in the next one.
— Vivek
P.S. Know a developer or founder trying to control their AI spend without giving up on agentic features? Forward this to them. They can subscribe at https://savvymonk.beehiiv.com/