Claude Sonnet 5 is live on AF: 63.2% on agentic coding, at four-tenths the price of Opus 4.8
Author
AgentsFlare Research
Date Published
If you read nothing else this week.
- Takeaway #1: AF has added Claude Sonnet 5, Anthropic’s new mid-tier model for agentic coding, tool use, and long-horizon autonomous tasks, replacing Sonnet 4.6.
- Takeaway #2: Sonnet 5 moves much closer to flagship-level performance: 63.2% on SWE-bench Pro and 80.4% on Terminal-Bench 2.1, making it suitable for multi-file code changes, cross-tool research, and long-chain automation.
- Takeaway #3: Pricing is the main advantage: $2/$10 during the promotion period and $3/$15 at standard pricing, roughly 40% of Opus 4.8’s cost, with a 1M-token context window included.
- Takeaway #4: On AF, enterprises can use Sonnet 5 as the main model for long-chain tasks, reserve Opus 4.8 for the hardest reasoning steps, route simple classification to smaller models, and control spend through project-level budgets, quotas, and permissions.
Claude Sonnet 5, released by Anthropic on June 30, is now live on AgentsFlare, replacing the previous Sonnet 4.6.
As the new generation of Anthropic's mid-tier line, it can plan on its own, call tools like browsers and terminals, and keep running autonomously across longer task chains—work that just a few months ago took bigger, pricier models to handle. Anthropic has pushed its capability close to the flagship Opus 4.8 while keeping the price in mid-tier territory.
On pricing, the introductory rate is $2 input / $10 output per million tokens through August 31, after which it returns to the standard $3 / $15. The context window is 1M tokens, included at standard pricing with no long-context surcharge. Training data runs through January 2026.
This generation replaces February's Sonnet 4.6. On agentic coding (SWE-bench Pro) it moves from 4.6's 58.1% to 63.2%; on Terminal-Bench 2.1, which stresses multi-step tool use, it jumps from 67.0% to 80.4%—long-horizon autonomy and tool use are where the gains show most. Opus 4.8 scores 69.2% on the same agentic-coding test, so Sonnet 5 reaches roughly 90% of the flagship's result at a mid-tier price. Against the comparable options from Google and OpenAI—Gemini 3 Pro at $2 / $12 per million tokens (up to 200K), GPT-5.5 at $5 / $30—Sonnet 5's introductory $2 / $10 is cheaper on the output side, and its standard $3 / $15 still sits below both flagships.
Where it fits: work where the model has to run a long string of steps on its own—multi-file changes across a codebase, cross-tool research and synthesis, automation flows that chain terminal or browser calls. Take a support-ticket automation project where the model reads the ticket, checks internal docs, calls APIs, and writes a reply, tens of thousands of calls a day: Opus 4.8 is hard to justify on cost, a smaller model can't hold the multi-step reasoning, and Sonnet 5 lands right in between—cutting unit cost while holding completion rate.
Price and billing comparison
* table content*
Model
Input / Output (per 1M tokens)
Context
Agentic coding (SWE-bench Pro)
Notes
Claude Sonnet 5
$2 / $10 (intro, through 8/31); standard $3 / $15
1M
63.2%
This release, live on AgentsFlare
Claude Opus 4.8
$5 / $25
1M
69.2%
Anthropic flagship, capability anchor
Gemini 3 Pro
$2 / $12 (≤200K; above: $4 / $18)
—
—
Google peer
GPT-5.5
$5 / $30
1M
—
OpenAI flagship
All prices per each vendor's official site. Sonnet 5's $2 / $10 is a limited-time introductory rate that switches to the standard $3 / $15 after August 31.
Using it on AgentsFlare
Model choice stays in your hands. On AgentsFlare you set up projects by your own rules, make Sonnet 5 the workhorse for long-chain tasks, keep Opus 4.8 for the hardest reasoning steps and a smaller model for simple classification, and define the whole call order yourself. Switch freely between APIs without touching your business code. With project management on top, you can set budget, quota, and permissions per project—give the high-volume Sonnet 5 automation project its own ledger and keep cost and usage in clear view.
Getting started
Sonnet 5 is callable on AgentsFlare now, with the introductory price window open through August 31. Teams looking to cut cost on long-horizon agent tasks can wire it in and try it. For model pairing and budget planning around a specific use case, reach out to the AgentsFlare team.