Claude Opus 5.5
Long coding and knowledge work
$4 / $20 input / output · per 1M tokens
USD · standard API text, base input / output per 1M tokens
Try a reviewable change; record retries and corrections.
Provider source (opens in a new tab)Choose a route for the job, not a winner for everyone.
Nine current model starting points, official API terms and a small, independently measured snapshot. Facts checked September 25, 2026; these are not our head-to-head tests.
In focusSonnet 5.5 · released Sep 28 →GPT-6 Sol + Luna · official release →Opus 5.5 · launch & reactions →Jev · what people are building →Free access & alternatives →
If you have a long coding task
Start with a reviewable finish.
Opus 5.5 and GPT-6 Sol are both offered for coding and agent work. Try one bounded task in a product you already use. Compare accepted changes, failed attempts and correction time before adding another plan.
Use the complete free decision guide →If you have many small tasks
Price the whole workflow.
GPT-6 Luna has a much lower published text-token rate, but an API price does not describe a subscription allowance, tool calls or human review. A narrow extraction or classification job is a more meaningful first test than a broad design brief.
Read the official release notebook →Official provider snapshot / September 25, 2026
Nine models. Different conditions.
Published API text prices per one million input / output tokens. These are not ChatGPT, Claude or Codex subscription prices. Currency, context tier, caching and time-of-day conditions are shown on each row; tools may add cost.
| Model | Published task fit | API input / output | Condition and a useful first check |
|---|---|---|---|
| Claude Opus 5.5Anthropic | Long coding and knowledge work Provider source (opens in a new tab) | $4 / $20 | USD · standard API text, base input / output per 1M tokens Try a reviewable change; record retries and corrections. |
| Claude Fable 5.1Anthropic | Demanding reasoning and long-horizon work Provider source (opens in a new tab) | $10 / $50 | USD · standard API text, base input / output per 1M tokens Start with Opus unless a task-specific evaluation justifies the added cost. |
| GPT-6 AstraOpenAI | Difficult agent work and deep review Provider source (opens in a new tab) | $10 / $50 | USD · Standard, short-context API text per 1M input / output tokens Compare accepted changes and correction time, not only first-answer quality. |
| GPT-6 SolOpenAI | Complex coding and agent workflows Provider source (opens in a new tab) | $2 / $10 | USD · Standard, short-context API text per 1M input / output tokens Check whether a bounded task reaches a complete, reviewable finish. |
| GPT-6 LunaOpenAI | Focused, repeatable, high-volume work Provider source (opens in a new tab) | $0.10 / $0.50 | USD · Standard, short-context API text per 1M input / output tokens Try a narrow extraction or classification job with a quality check. |
| DeepSeek V4.1 FlashDeepSeek | Low-cost text and vision experimentation Provider source (opens in a new tab) | $0.30 / $1.20 | USD · peak, cache-miss input / output per 1M tokens; off-peak is lower Keep thinking mode and peak/off-peak billing visible in a real-task comparison. |
| GLM-5.1Z.AI | Coding and reasoning candidate Provider source (opens in a new tab) | $1.40 / $4.40 | USD · API uncached input / output per 1M tokens Verify availability, language and total task cost in your own route. |
| Kimi K3Moonshot AI | Software engineering and deep reasoning candidate Provider source (opens in a new tab) | ¥20 / ¥100 | CNY · API input / output per 1M tokens; not converted to USD Compare a complete task; do not rank this CNY quote as a USD price. |
| Grok 4.7xAI | Coding and agentic work candidate Provider source (opens in a new tab) | $2 / $6 | USD · API input / output per 1M tokens below 200K prompt tokens; higher above Confirm the context tier and tool use before estimating a full run. |
Claude Fable 5.1
Demanding reasoning and long-horizon work
$10 / $50 input / output · per 1M tokens
USD · standard API text, base input / output per 1M tokens
Start with Opus unless a task-specific evaluation justifies the added cost.
Provider source (opens in a new tab)GPT-6 Astra
Difficult agent work and deep review
$10 / $50 input / output · per 1M tokens
USD · Standard, short-context API text per 1M input / output tokens
Compare accepted changes and correction time, not only first-answer quality.
Provider source (opens in a new tab)GPT-6 Sol
Complex coding and agent workflows
$2 / $10 input / output · per 1M tokens
USD · Standard, short-context API text per 1M input / output tokens
Check whether a bounded task reaches a complete, reviewable finish.
Provider source (opens in a new tab)GPT-6 Luna
Focused, repeatable, high-volume work
$0.10 / $0.50 input / output · per 1M tokens
USD · Standard, short-context API text per 1M input / output tokens
Try a narrow extraction or classification job with a quality check.
Provider source (opens in a new tab)DeepSeek V4.1 Flash
Low-cost text and vision experimentation
$0.30 / $1.20 input / output · per 1M tokens
USD · peak, cache-miss input / output per 1M tokens; off-peak is lower
Keep thinking mode and peak/off-peak billing visible in a real-task comparison.
Provider source (opens in a new tab)GLM-5.1
Coding and reasoning candidate
$1.40 / $4.40 input / output · per 1M tokens
USD · API uncached input / output per 1M tokens
Verify availability, language and total task cost in your own route.
Provider source (opens in a new tab)Kimi K3
Software engineering and deep reasoning candidate
¥20 / ¥100 input / output · per 1M tokens
CNY · API input / output per 1M tokens; not converted to USD
Compare a complete task; do not rank this CNY quote as a USD price.
Provider source (opens in a new tab)Grok 4.7
Coding and agentic work candidate
$2 / $6 input / output · per 1M tokens
USD · API input / output per 1M tokens below 200K prompt tokens; higher above
Confirm the context tier and tool use before estimating a full run.
Provider source (opens in a new tab)Independent measurement / one method
A score is a question, not a purchase decision.
Selected Artificial Analysis Intelligence Index v4.3.2 results, checked September 25; Sonnet 5.5 added September 29. The index combines ten evaluations. Effort and fallback settings are part of each result. We show a small attributed excerpt, not a full leaderboard or our own testing.
Claude Opus 5.5adaptive · max effort · default fallback
58Measurement & method (opens in a new tab)Claude Sonnet 5.5adaptive · max effort · default fallback
56Measurement & method (opens in a new tab)GPT-6 Solmax effort
48Measurement & method (opens in a new tab)GPT-6 Solhigh effort
43Measurement & method (opens in a new tab)DeepSeek V4.1 Flashreasoning · max effort
39Measurement & method (opens in a new tab)Different modes can change both results and task cost. Missing models are not scored zero; inspect the publisher’s current data and run your own representative task. No single index measures design taste, subscription limits, privacy or correction effort.
Product layer / not the model table
What the price column cannot answer.
Subscription
Confirm your current plan’s availability, usage limits and resets in that product. Do not convert an API token price into an unlimited monthly allowance.
Toolchain
Images, voice, repository access and connected software can involve different systems. A model’s text price cannot establish the result or rights of an image or video workflow.
Our editorial recommendation: compare the finished artifact, correction effort, permissions and total task cost. We have not run matched tasks across these models.
Keep a benchmark, a vendor claim and a creator report separate.
Provider prices come from official rate sheets. The four scored observations above come from one independent publisher with explicit settings. Practitioner stories are sourced experiences, not controlled comparisons. We do not reproduce the publisher’s full chart or rank the nine models by one number.
Inspect the publisher’s original result (opens in a new tab)The public editionGood ideas deserve
Explore Jev for free ↗Good ideas deserve
an open door.
Explore the Jev collection, inspect the creators’ work and choose a useful next experiment. No subscription required.